diff --git a/src/libcore/tests/time.rs b/src/libcore/tests/time.rs
index b0abdc749f6..ef3e1acfbf5 100644
--- a/src/libcore/tests/time.rs
+++ b/src/libcore/tests/time.rs
@@ -219,11 +219,8 @@ fn debug_formatting_precision_zero() {
 
 #[test]
 fn debug_formatting_precision_two() {
-    // This might seem inconsistent with the other units, but printing
-    // fractional digits for nano seconds would imply more precision than is
-    // actually stored.
-    assert_eq!(format!("{:.2?}", Duration::new(0, 0)), "0ns");
-    assert_eq!(format!("{:.2?}", Duration::new(0, 123)), "123ns");
+    assert_eq!(format!("{:.2?}", Duration::new(0, 0)), "0.00ns");
+    assert_eq!(format!("{:.2?}", Duration::new(0, 123)), "123.00ns");
 
     assert_eq!(format!("{:.2?}", Duration::new(0, 1_000)), "1.00µs");
     assert_eq!(format!("{:.2?}", Duration::new(0, 7_001)), "7.00µs");
@@ -244,3 +241,12 @@ fn debug_formatting_precision_two() {
     assert_eq!(format!("{:.2?}", Duration::new(2, 105_000_000)), "2.11s");
     assert_eq!(format!("{:.2?}", Duration::new(8, 999_999_999)), "9.00s");
 }
+
+#[test]
+fn debug_formatting_precision_high() {
+    assert_eq!(format!("{:.5?}",  Duration::new(0, 23_678)), "23.67800µs");
+
+    assert_eq!(format!("{:.9?}",  Duration::new(1, 000_000_000)), "1.000000000s");
+    assert_eq!(format!("{:.10?}", Duration::new(4, 001_000_000)), "4.0010000000s");
+    assert_eq!(format!("{:.20?}", Duration::new(4, 001_000_000)), "4.00100000000000000000s");
+}
diff --git a/src/libcore/time.rs b/src/libcore/time.rs
index 34bf3637f29..9703c61fe92 100644
--- a/src/libcore/time.rs
+++ b/src/libcore/time.rs
@@ -511,9 +511,9 @@ impl fmt::Debug for Duration {
             // The next digit is written at this position
             let mut pos = 0;
 
-            // We can stop when there are no non-zero digits left or (when a
-            // precision was set and we already emitted that many digits).
-            while fractional_part > 0 && f.precision().map(|p| p > pos).unwrap_or(true) {
+            // We keep writing digits into the buffer while there are non-zero
+            // digits left and we haven't written enough digits yet.
+            while fractional_part > 0 && pos < f.precision().unwrap_or(9) {
                 // Write new digit into the buffer
                 buf[pos] = b'0' + (fractional_part / divisor) as u8;
 
@@ -556,9 +556,13 @@ impl fmt::Debug for Duration {
                 }
             }
 
+            // Determine the end of the buffer: if precision is set, we just
+            // use as many digits from the buffer (capped to 9). If it isn't
+            // set, we only use all digits up to the last non-zero one.
+            let end = f.precision().map(|p| ::cmp::min(p, 9)).unwrap_or(pos);
+
             // If we haven't emitted a single fractional digit and the precision
             // wasn't set to a non-zero value, we don't print the decimal point.
-            let end = f.precision().unwrap_or(pos);
             if end == 0 {
                 write!(f, "{}", integer_part)
             } else {
@@ -568,7 +572,9 @@ impl fmt::Debug for Duration {
                     ::str::from_utf8_unchecked(&buf[..end])
                 };
 
-                write!(f, "{}.{}", integer_part, s)
+                // If the user request a precision > 9, we pad '0's at the end.
+                let w = f.precision().unwrap_or(pos);
+                write!(f, "{}.{:0<width$}", integer_part, s, width = w)
             }
         }
 
@@ -587,7 +593,8 @@ impl fmt::Debug for Duration {
             fmt_decimal(f, self.nanos as u64 / 1_000, self.nanos % 1_000, 100)?;
             f.write_str("µs")
         } else {
-            write!(f, "{}ns", self.nanos)
+            fmt_decimal(f, self.nanos as u64, 0, 1)?;
+            f.write_str("ns")
         }
     }
 }