diff --git a/CHANGELOG.md b/CHANGELOG.md
index de12bd4..d65999b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -17,6 +17,8 @@ This project attempts to adhere to [Semantic Versioning](http://semver.org).
   - reduce runtime in our own benchmark by more than `70%`
   - reduce binary size by more than `25%`
 
+- only boost similarity in Jaro-Winkler once the Jaro similarity exceeds 0.7
+
 ### Fixed
 
 - Fix transposition counting in Jaro and Jaro-Winkler.
diff --git a/src/lib.rs b/src/lib.rs
index 0d0c78c..f761e3e 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -179,16 +179,20 @@ where
     &'b Iter2: IntoIterator<Item = Elem2>,
     Elem1: PartialEq<Elem2>,
 {
-    let jaro_distance = generic_jaro(a, b);
+    let sim = generic_jaro(a, b);
 
-    let prefix_length = a
-        .into_iter()
-        .take(4)
-        .zip(b.into_iter())
-        .take_while(|(a_elem, b_elem)| a_elem == b_elem)
-        .count();
+    if sim > 0.7 {
+        let prefix_length = a
+            .into_iter()
+            .take(4)
+            .zip(b.into_iter())
+            .take_while(|(a_elem, b_elem)| a_elem == b_elem)
+            .count();
 
-    return jaro_distance + 0.1 * prefix_length as f64 * (1.0 - jaro_distance);
+        sim + 0.1 * prefix_length as f64 * (1.0 - sim)
+    } else {
+        sim
+    }
 }
 
 /// Like Jaro but gives a boost to strings that have a common prefix.
@@ -918,7 +922,7 @@ mod tests {
 
     #[test]
     fn jaro_winkler_names() {
-        assert!((0.562 - jaro_winkler("Friedrich Nietzsche", "Fran-Paul Sartre")).abs() < 0.001);
+        assert!((0.452 - jaro_winkler("Friedrich Nietzsche", "Fran-Paul Sartre")).abs() < 0.001);
     }
 
     #[test]