diff --git a/leaf/src/app/outbound/manager.rs b/leaf/src/app/outbound/manager.rs
index 14fe6f574..555c6ce93 100644
--- a/leaf/src/app/outbound/manager.rs
+++ b/leaf/src/app/outbound/manager.rs
@@ -465,6 +465,8 @@ impl OutboundManager {
                             settings.health_check_prefers.clone(),
                             settings.health_check_on_start,
                             settings.health_check_wait,
+                            settings.health_check_attempts,
+                            settings.health_check_success_percentage,
                             dns_client.clone(),
                         );
                         let (datagram, mut datagram_abort_handles) = failover::DatagramHandler::new(
@@ -480,6 +482,8 @@ impl OutboundManager {
                             settings.health_check_prefers,
                             settings.health_check_on_start,
                             settings.health_check_wait,
+                            settings.health_check_attempts,
+                            settings.health_check_success_percentage,
                             dns_client.clone(),
                         );
                         let handler = HandlerBuilder::default()
diff --git a/leaf/src/config/conf/config.rs b/leaf/src/config/conf/config.rs
index fc09fb608..39bc5c0a0 100644
--- a/leaf/src/config/conf/config.rs
+++ b/leaf/src/config/conf/config.rs
@@ -135,6 +135,8 @@ pub struct ProxyGroup {
     pub health_check_prefers: Option<Vec<String>>,
     pub health_check_on_start: Option<bool>,
     pub health_check_wait: Option<bool>,
+    pub health_check_attempts: Option<u32>,
+    pub health_check_success_percentage: Option<u32>,
 
     // tryall
     pub delay_base: Option<u32>,
@@ -163,6 +165,8 @@ impl Default for ProxyGroup {
             health_check_prefers: None,
             health_check_on_start: None,
             health_check_wait: None,
+            health_check_attempts: None,
+            health_check_success_percentage: None,
             delay_base: None,
             method: None,
         }
@@ -626,6 +630,14 @@ pub fn from_lines(lines: Vec<io::Result<String>>) -> Result<Config> {
                         group.health_check_wait =
                             if v == "true" { Some(true) } else { Some(false) };
                     }
+                    "health-check-attempts" => {
+                        let i = if let Ok(i) = v.parse() { Some(i) } else { None };
+                        group.health_check_attempts = i;
+                    }
+                    "health-check-success-percentage" => {
+                        let i = if let Ok(i) = v.parse() { Some(i) } else { None };
+                        group.health_check_success_percentage = i;
+                    }
                     "delay-base" => {
                         let i = if let Ok(i) = v.parse() { Some(i) } else { None };
                         group.delay_base = i;
@@ -1317,6 +1329,11 @@ pub fn to_internal(conf: &mut Config) -> Result<internal::Config> {
                     settings.health_check_on_start =
                         ext_proxy_group.health_check_on_start.unwrap_or(false);
                     settings.health_check_wait = ext_proxy_group.health_check_wait.unwrap_or(false);
+                    settings.health_check_attempts =
+                        ext_proxy_group.health_check_attempts.unwrap_or(1);
+                    settings.health_check_success_percentage = ext_proxy_group
+                        .health_check_success_percentage
+                        .unwrap_or(50);
                     let settings = settings.write_to_bytes().unwrap();
                     outbound.settings = settings;
                     outbounds.push(outbound);
diff --git a/leaf/src/config/internal/config.proto b/leaf/src/config/internal/config.proto
index 6abb20f44..b61e9af1f 100644
--- a/leaf/src/config/internal/config.proto
+++ b/leaf/src/config/internal/config.proto
@@ -172,6 +172,7 @@ message ChainOutboundSettings {
 }
 
 message FailOverOutboundSettings {
+  // A list of outbound tags.
 	repeated string actors = 1;
   // The connect timeout value for the outbounds, in seconds. Some outbounds don't
   // need to "connect", those outbounds have zero connect time. Default 4.
@@ -186,6 +187,7 @@ message FailOverOutboundSettings {
 	bool fallback_cache = 6;
 	uint32 cache_size = 7;
 	uint32 cache_timeout = 8;
+  // The outbound to be used when all actors are unavailable.
 	optional string last_resort = 9;
   // Timeout for health check request, in seconds. Default is 6.
 	uint32 health_check_timeout = 10;
@@ -204,6 +206,14 @@ message FailOverOutboundSettings {
   // Hold incomming connections, wait for the initial health check to be done.
   // Default false.
 	bool health_check_wait = 15;
+  // Total number of health check requests per outbound, the RTT to be used to sort
+  // the outbounds is the average of all successful attempts. Default 1.
+	uint32 health_check_attempts = 16;
+  // The percentage of successful health check requests which to consider an outbound
+  // is available, must be a value between [0, 100], if the result is lower than this
+  // percentage, the outbound's RTT would be set to a timeout value, thus marks the
+  // outbound as unavailable. Default 50.
+	uint32 health_check_success_percentage = 17;
 }
 
 message SelectOutboundSettings {
diff --git a/leaf/src/config/internal/config.rs b/leaf/src/config/internal/config.rs
index 10059e382..d15e41a25 100644
--- a/leaf/src/config/internal/config.rs
+++ b/leaf/src/config/internal/config.rs
@@ -3222,6 +3222,10 @@ pub struct FailOverOutboundSettings {
     pub health_check_on_start: bool,
     // @@protoc_insertion_point(field:FailOverOutboundSettings.health_check_wait)
     pub health_check_wait: bool,
+    // @@protoc_insertion_point(field:FailOverOutboundSettings.health_check_attempts)
+    pub health_check_attempts: u32,
+    // @@protoc_insertion_point(field:FailOverOutboundSettings.health_check_success_percentage)
+    pub health_check_success_percentage: u32,
     // special fields
     // @@protoc_insertion_point(special_field:FailOverOutboundSettings.special_fields)
     pub special_fields: ::protobuf::SpecialFields,
@@ -3294,6 +3298,12 @@ impl ::protobuf::Message for FailOverOutboundSettings {
                 120 => {
                     self.health_check_wait = is.read_bool()?;
                 },
+                128 => {
+                    self.health_check_attempts = is.read_uint32()?;
+                },
+                136 => {
+                    self.health_check_success_percentage = is.read_uint32()?;
+                },
                 tag => {
                     ::protobuf::rt::read_unknown_or_skip_group(tag, is, self.special_fields.mut_unknown_fields())?;
                 },
@@ -3351,6 +3361,12 @@ impl ::protobuf::Message for FailOverOutboundSettings {
         if self.health_check_wait != false {
             my_size += 1 + 1;
         }
+        if self.health_check_attempts != 0 {
+            my_size += ::protobuf::rt::uint32_size(16, self.health_check_attempts);
+        }
+        if self.health_check_success_percentage != 0 {
+            my_size += ::protobuf::rt::uint32_size(17, self.health_check_success_percentage);
+        }
         my_size += ::protobuf::rt::unknown_fields_size(self.special_fields.unknown_fields());
         self.special_fields.cached_size().set(my_size as u32);
         my_size
@@ -3402,6 +3418,12 @@ impl ::protobuf::Message for FailOverOutboundSettings {
         if self.health_check_wait != false {
             os.write_bool(15, self.health_check_wait)?;
         }
+        if self.health_check_attempts != 0 {
+            os.write_uint32(16, self.health_check_attempts)?;
+        }
+        if self.health_check_success_percentage != 0 {
+            os.write_uint32(17, self.health_check_success_percentage)?;
+        }
         os.write_unknown_fields(self.special_fields.unknown_fields())?;
         ::std::result::Result::Ok(())
     }
@@ -3434,6 +3456,8 @@ impl ::protobuf::Message for FailOverOutboundSettings {
         self.health_check_prefers.clear();
         self.health_check_on_start = false;
         self.health_check_wait = false;
+        self.health_check_attempts = 0;
+        self.health_check_success_percentage = 0;
         self.special_fields.clear();
     }
 
@@ -3454,6 +3478,8 @@ impl ::protobuf::Message for FailOverOutboundSettings {
             health_check_prefers: ::std::vec::Vec::new(),
             health_check_on_start: false,
             health_check_wait: false,
+            health_check_attempts: 0,
+            health_check_success_percentage: 0,
             special_fields: ::protobuf::SpecialFields::new(),
         };
         &instance
diff --git a/leaf/src/config/json/config.rs b/leaf/src/config/json/config.rs
index 6f7a035ad..dd581ddd2 100644
--- a/leaf/src/config/json/config.rs
+++ b/leaf/src/config/json/config.rs
@@ -218,6 +218,10 @@ pub struct FailOverOutboundSettings {
     pub health_check_on_start: Option<bool>,
     #[serde(rename = "healthCheckWait")]
     pub health_check_wait: Option<bool>,
+    #[serde(rename = "healthCheckAttempts")]
+    pub health_check_attempts: Option<u32>,
+    #[serde(rename = "healthCheckSuccessPercentage")]
+    pub health_check_success_percentage: Option<u32>,
     pub failover: Option<bool>,
     #[serde(rename = "fallbackCache")]
     pub fallback_cache: Option<bool>,
@@ -810,6 +814,10 @@ pub fn to_internal(json: &mut Config) -> Result<internal::Config> {
                     settings.health_check_on_start =
                         ext_settings.health_check_on_start.unwrap_or(false);
                     settings.health_check_wait = ext_settings.health_check_wait.unwrap_or(false);
+                    settings.health_check_attempts =
+                        ext_settings.health_check_attempts.unwrap_or(1);
+                    settings.health_check_success_percentage =
+                        ext_settings.health_check_success_percentage.unwrap_or(50);
                     settings.check_interval = ext_settings.check_interval.unwrap_or(300); // 300 secs
                     settings.failover = ext_settings.failover.unwrap_or(true);
                     settings.fallback_cache = ext_settings.fallback_cache.unwrap_or(false);
diff --git a/leaf/src/proxy/failover/datagram.rs b/leaf/src/proxy/failover/datagram.rs
index 6bb2ce846..65290656d 100644
--- a/leaf/src/proxy/failover/datagram.rs
+++ b/leaf/src/proxy/failover/datagram.rs
@@ -38,6 +38,8 @@ impl Handler {
         health_check_prefers: Vec<String>,
         health_check_on_start: bool,
         health_check_wait: bool,
+        health_check_attempts: u32,
+        health_check_success_percentage: u32,
         dns_client: SyncDnsClient,
     ) -> (Self, Vec<AbortHandle>) {
         let mut abort_handles = Vec::new();
@@ -66,6 +68,8 @@ impl Handler {
                 last_active.clone(),
                 is_first_health_check_done.clone(),
                 notify.as_ref().cloned(),
+                health_check_attempts,
+                health_check_success_percentage,
             ));
             abort_handles.push(abort_handle);
             let task: BoxFuture<'static, ()> = Box::pin(abortable.map(|_| ()));
diff --git a/leaf/src/proxy/failover/mod.rs b/leaf/src/proxy/failover/mod.rs
index 126268315..e939e4545 100644
--- a/leaf/src/proxy/failover/mod.rs
+++ b/leaf/src/proxy/failover/mod.rs
@@ -40,9 +40,12 @@ async fn single_health_check(
     tag: String,
     h: AnyOutboundHandler,
     dns_client: SyncDnsClient,
-    delay: Duration,
+    delay: u32,
 ) -> Measure {
-    tokio::time::sleep(delay).await;
+    tokio::time::sleep(Duration::from_millis(
+        StdRng::from_entropy().gen_range(0..=delay) as u64,
+    ))
+    .await;
 
     let dest = match network {
         Network::Tcp => SocksAddr::Domain("www.google.com".to_string(), 443),
@@ -91,12 +94,14 @@ async fn single_health_check(
                     let mut buf = BytesMut::with_capacity(2 * 1024);
                     match stream.read_buf(&mut buf).await {
                         Ok(n) => {
+                            let elapsed = Instant::now().duration_since(start);
                             debug!(
-                                "received {} bytes tcp health check response: {}",
+                                "received {} bytes tcp health check response from {} in {} ms: {}",
                                 n,
+                                &tag,
+                                elapsed.as_millis(),
                                 String::from_utf8_lossy(&buf[..n.min(12)]),
                             );
-                            let elapsed = Instant::now().duration_since(start);
                             m = Measure::new(idx, elapsed.as_millis(), tag);
                         }
                         Err(_) => {
@@ -158,8 +163,13 @@ async fn single_health_check(
                     let mut buf = vec![0u8; 1500];
                     match recv.recv_from(&mut buf).await {
                         Ok((n, _)) => {
-                            debug!("received {} bytes udp health check response", n);
                             let elapsed = tokio::time::Instant::now().duration_since(start);
+                            debug!(
+                                "received {} bytes udp health check response from {} in {} ms",
+                                n,
+                                &tag,
+                                elapsed.as_millis()
+                            );
                             Measure::new(idx, elapsed.as_millis(), tag)
                         }
                         Err(_) => Measure::new(idx, u128::MAX - 3, tag),
@@ -177,17 +187,58 @@ async fn health_check(
     tag: String,
     h: AnyOutboundHandler,
     dns_client: SyncDnsClient,
-    delay: Duration,
+    delay: u32,
     health_check_timeout: u64,
+    health_check_attempts: u32,
+    health_check_success_percentage: u32,
 ) -> Measure {
     debug!("health checking [{}] ({}) index ({})", &tag, &network, idx);
+    let health_check_timeout = Duration::from_secs(health_check_timeout);
+    let health_check_timeout_ms = health_check_timeout.as_millis();
+    let mut attempts = Vec::new();
+    for _ in 0..health_check_attempts {
+        attempts.push(timeout(
+            health_check_timeout,
+            single_health_check(
+                network,
+                idx,
+                tag.clone(),
+                h.clone(),
+                dns_client.clone(),
+                delay,
+            ),
+        ));
+    }
+    let measures = futures::future::join_all(attempts).await;
+    let measures = measures
+        .into_iter()
+        .map(|x| x.unwrap_or(Measure::new(idx, u128::MAX - 1, tag.clone())))
+        .collect::<Vec<_>>();
+
+    let n_success = measures
+        .iter()
+        .filter(|x| x.rtt < health_check_timeout_ms as u128)
+        .count();
+
+    debug!(
+        "{} out of {} successful checks for {} [{}]",
+        n_success, health_check_attempts, &network, &tag
+    );
+
+    use std::ops::Div;
+    let success_percentage = ((n_success as f32).div(health_check_attempts as f32) * 100.) as u32;
+    if success_percentage < health_check_success_percentage {
+        return Measure::new(idx, u128::MAX, tag);
+    }
+
+    let mean_rtt = measures
+        .iter()
+        .filter(|x| x.rtt < health_check_timeout_ms as u128)
+        .map(|x| x.rtt)
+        .sum::<u128>()
+        .div_euclid(n_success as u128);
 
-    timeout(
-        Duration::from_secs(health_check_timeout),
-        single_health_check(network, idx, tag.clone(), h, dns_client, delay),
-    )
-    .await
-    .unwrap_or(Measure::new(idx, u128::MAX - 1, tag))
+    Measure::new(idx, mean_rtt, tag)
 }
 
 pub(self) async fn health_check_task(
@@ -205,6 +256,8 @@ pub(self) async fn health_check_task(
     last_active: Arc<Mutex<Instant>>,
     is_first_health_check_done: Arc<AtomicBool>,
     wait_for_health_check: Option<Arc<Notify>>,
+    health_check_attempts: u32,
+    health_check_success_percentage: u32,
 ) {
     loop {
         let last_active = Instant::now()
@@ -213,18 +266,18 @@ pub(self) async fn health_check_task(
 
         if last_active < health_check_active.into() {
             let mut checks = Vec::new();
-            let mut rng = StdRng::from_entropy();
             for (i, a) in (&actors).iter().enumerate() {
                 let dns_client_cloned = dns_client.clone();
-                let delay = Duration::from_millis(rng.gen_range(0..=health_check_delay) as u64);
                 checks.push(Box::pin(health_check(
                     network,
                     i,
                     a.tag().to_owned(),
                     a.clone(),
                     dns_client_cloned,
-                    delay,
+                    health_check_delay,
                     health_check_timeout as u64,
+                    health_check_attempts,
+                    health_check_success_percentage,
                 )));
             }
             let mut measures = futures::future::join_all(checks).await;
diff --git a/leaf/src/proxy/failover/stream.rs b/leaf/src/proxy/failover/stream.rs
index 8671b2891..6dc095628 100644
--- a/leaf/src/proxy/failover/stream.rs
+++ b/leaf/src/proxy/failover/stream.rs
@@ -44,6 +44,8 @@ impl Handler {
         health_check_prefers: Vec<String>,
         health_check_on_start: bool,
         health_check_wait: bool,
+        health_check_attempts: u32,
+        health_check_success_percentage: u32,
         dns_client: SyncDnsClient,
     ) -> (Self, Vec<AbortHandle>) {
         let mut abort_handles = Vec::new();
@@ -72,6 +74,8 @@ impl Handler {
                 last_active.clone(),
                 is_first_health_check_done.clone(),
                 notify.as_ref().cloned(),
+                health_check_attempts,
+                health_check_success_percentage,
             ));
             abort_handles.push(abort_handle);
             let task: BoxFuture<'static, ()> = Box::pin(abortable.map(|_| ()));