diff --git a/boa/src/builtins/number/mod.rs b/boa/src/builtins/number/mod.rs
index e1753997bc..d973baba4b 100644
--- a/boa/src/builtins/number/mod.rs
+++ b/boa/src/builtins/number/mod.rs
@@ -318,28 +318,57 @@ impl Number {
     ///   represented by these digits is rounded using string
     ///   manipulation.
     /// - Else, zeroes are appended to the string.
+    /// - Additionnally, sometimes the exponent was wrongly computed and
+    ///   while up-rounding we find that we need an extra digit. When this
+    ///   happens, we return true so that the calling context can adjust
+    ///   the exponent. The string is kept at an exact length of `precision`.
     ///
     /// When this procedure returns, `digits` is exactly `precision` long.
     ///
-    fn round_to_precision(digits: &mut String, precision: usize) {
+    fn round_to_precision(digits: &mut String, precision: usize) -> bool {
         if digits.len() > precision {
             let to_round = digits.split_off(precision);
             let mut digit = digits.pop().unwrap() as u8;
+            if let Some(first) = to_round.chars().next() {
+                if first > '4' {
+                    digit += 1;
+                }
+            }
 
-            for c in to_round.chars() {
-                match c {
-                    c if c < '4' => break,
-                    c if c > '4' => {
-                        digit += 1;
-                        break;
+            if digit as char == ':' {
+                // ':' is '9' + 1
+                // need to propagate the increment backward
+                let mut replacement = String::from("0");
+                let mut propagated = false;
+                for c in digits.chars().rev() {
+                    let d = match (c, propagated) {
+                        ('0'..='8', false) => (c as u8 + 1) as char,
+                        (_, false) => '0',
+                        (_, true) => c,
+                    };
+                    replacement.push(d);
+                    if d != '0' {
+                        propagated = true;
                     }
-                    _ => {}
                 }
+                digits.clear();
+                let replacement = if !propagated {
+                    digits.push('1');
+                    &replacement.as_str()[1..]
+                } else {
+                    replacement.as_str()
+                };
+                for c in replacement.chars().rev() {
+                    digits.push(c)
+                }
+                !propagated
+            } else {
+                digits.push(digit as char);
+                false
             }
-
-            digits.push(digit as char);
         } else {
             digits.push_str(&"0".repeat(precision - digits.len()));
+            false
         }
     }
 
@@ -359,17 +388,24 @@ impl Number {
         args: &[Value],
         context: &mut Context,
     ) -> Result<Value> {
-        let precision_var = args.get(0).cloned().unwrap_or_default();
+        let precision = args.get(0).cloned().unwrap_or_default();
 
         // 1 & 6
         let mut this_num = Self::this_number_value(this, context)?;
-        // 2 & 4
-        if precision_var == Value::undefined() || !this_num.is_finite() {
+        // 2
+        if precision == Value::undefined() {
             return Self::to_string(this, &[], context);
         }
 
         // 3
-        let precision = match precision_var.to_integer_or_infinity(context)? {
+        let precision = precision.to_integer_or_infinity(context)?;
+
+        // 4
+        if !this_num.is_finite() {
+            return Self::to_string(this, &[], context);
+        }
+
+        let precision = match precision {
             IntegerOrInfinity::Integer(x) if (1..=100).contains(&x) => x as usize,
             _ => {
                 // 5
@@ -383,7 +419,7 @@ impl Number {
         // 7
         let mut prefix = String::new(); // spec: 's'
         let mut suffix: String; // spec: 'm'
-        let exponent: i32; // spec: 'e'
+        let mut exponent: i32; // spec: 'e'
 
         // 8
         if this_num < 0.0 {
@@ -399,10 +435,8 @@ impl Number {
         } else {
             // Due to f64 limitations, this part differs a bit from the spec,
             // but has the same effect. It manipulates the string constructed
-            // by ryu-js: digits with an optional dot between two of them.
-
-            let mut buffer = ryu_js::Buffer::new();
-            suffix = buffer.format(this_num).to_string();
+            // by `format`: digits with an optional dot between two of them.
+            suffix = format!("{:.100}", this_num);
 
             // a: getting an exponent
             exponent = Self::flt_str_to_exp(&suffix);
@@ -413,7 +447,9 @@ impl Number {
                 suffix.remove(n);
             }
             // impl: having exactly `precision` digits in `suffix`
-            Self::round_to_precision(&mut suffix, precision);
+            if Self::round_to_precision(&mut suffix, precision) {
+                exponent += 1;
+            }
 
             // c: switching to scientific notation
             let great_exp = exponent >= precision_i32;
diff --git a/boa/src/builtins/number/tests.rs b/boa/src/builtins/number/tests.rs
index e3f8d0ae50..634c527963 100644
--- a/boa/src/builtins/number/tests.rs
+++ b/boa/src/builtins/number/tests.rs
@@ -137,6 +137,8 @@ fn to_precision() {
         var exact_precision = (123456789).toPrecision(9);
         var over_precision = (123456789).toPrecision(50);
         var neg_precision = (-123456789).toPrecision(4);
+        var neg_exponent = (0.1).toPrecision(4);
+        var ieee754_limits = (1/3).toPrecision(60);
         "#;
 
     eprintln!("{}", forward(&mut context, init));
@@ -148,6 +150,8 @@ fn to_precision() {
     let exact_precision = forward(&mut context, "exact_precision");
     let over_precision = forward(&mut context, "over_precision");
     let neg_precision = forward(&mut context, "neg_precision");
+    let neg_exponent = forward(&mut context, "neg_exponent");
+    let ieee754_limits = forward(&mut context, "ieee754_limits");
 
     assert_eq!(infinity, String::from("\"Infinity\""));
     assert_eq!(default_precision, String::from("\"0\""));
@@ -160,6 +164,11 @@ fn to_precision() {
         over_precision,
         String::from("\"123456789.00000000000000000000000000000000000000000\"")
     );
+    assert_eq!(neg_exponent, String::from("\"0.1000\""));
+    assert_eq!(
+        ieee754_limits,
+        String::from("\"0.333333333333333314829616256247390992939472198486328125000000\"")
+    );
 
     let expected = "Uncaught \"RangeError\": \"precision must be an integer at least 1 and no greater than 100\"";
 
diff --git a/boa/src/vm/instructions.rs b/boa/src/vm/instructions.rs
index 39ea22abbd..0b40a2334c 100644
--- a/boa/src/vm/instructions.rs
+++ b/boa/src/vm/instructions.rs
@@ -12,7 +12,7 @@ pub enum Instruction {
     /// Loads an i32 onto the stack
     Int32(i32),
 
-    /// Loads an f32 onto the stack
+    /// Loads an f64 onto the stack
     Rational(f64),
 
     /// Adds the values from destination and source and stores the result in destination