diff --git a/include/fmt/format.h b/include/fmt/format.h
index 73686c40..9f312624 100644
--- a/include/fmt/format.h
+++ b/include/fmt/format.h
@@ -1373,34 +1373,26 @@ FMT_API auto fmt_snprintf(char* buf, size_t size, const char* fmt, ...) -> int;
 #  define FMT_SNPRINTF fmt_snprintf
 #endif  // _MSC_VER
 
-// Formats a floating-point number with snprintf.
+// Formats a floating-point number with snprintf using the hexfloat format.
 template <typename T>
 auto snprintf_float(T value, int precision, float_specs specs,
                     buffer<char>& buf) -> int {
   // Buffer capacity must be non-zero, otherwise MSVC's vsnprintf_s will fail.
   FMT_ASSERT(buf.capacity() > buf.size(), "empty buffer");
+  FMT_ASSERT(specs.format == float_format::hex, "");
   static_assert(!std::is_same<T, float>::value, "");
 
-  // Subtract 1 to account for the difference in precision since we use %e for
-  // both general and exponent format.
-  if (specs.format == float_format::general ||
-      specs.format == float_format::exp) {
-    precision = (precision >= 0 ? precision : 6) - 1;
-  }
-
   // Build the format string.
   char format[7];  // The longest format is "%#.*Le".
   char* format_ptr = format;
   *format_ptr++ = '%';
-  if (specs.showpoint && specs.format == float_format::hex) *format_ptr++ = '#';
+  if (specs.showpoint) *format_ptr++ = '#';
   if (precision >= 0) {
     *format_ptr++ = '.';
     *format_ptr++ = '*';
   }
   if (std::is_same<T, long double>()) *format_ptr++ = 'L';
-  *format_ptr++ = specs.format != float_format::hex
-                      ? (specs.format == float_format::fixed ? 'f' : 'e')
-                      : (specs.upper ? 'A' : 'a');
+  *format_ptr++ = specs.upper ? 'A' : 'a';
   *format_ptr = '\0';
 
   // Format using snprintf.
@@ -1422,55 +1414,11 @@ auto snprintf_float(T value, int precision, float_specs specs,
     }
     auto size = to_unsigned(result);
     // Size equal to capacity means that the last character was truncated.
-    if (size >= capacity) {
-      buf.try_reserve(size + offset + 1);  // Add 1 for the terminating '\0'.
-      continue;
-    }
-    auto is_digit = [](char c) { return c >= '0' && c <= '9'; };
-    if (specs.format == float_format::fixed) {
-      if (precision == 0) {
-        buf.try_resize(size);
-        return 0;
-      }
-      // Find and remove the decimal point.
-      auto end = begin + size, p = end;
-      do {
-        --p;
-      } while (is_digit(*p));
-      int fraction_size = static_cast<int>(end - p - 1);
-      std::memmove(p, p + 1, to_unsigned(fraction_size));
-      buf.try_resize(size - 1);
-      return -fraction_size;
-    }
-    if (specs.format == float_format::hex) {
+    if (size < capacity) {
       buf.try_resize(size + offset);
       return 0;
     }
-    // Find and parse the exponent.
-    auto end = begin + size, exp_pos = end;
-    do {
-      --exp_pos;
-    } while (*exp_pos != 'e');
-    char sign = exp_pos[1];
-    FMT_ASSERT(sign == '+' || sign == '-', "");
-    int exp = 0;
-    auto p = exp_pos + 2;  // Skip 'e' and sign.
-    do {
-      FMT_ASSERT(is_digit(*p), "");
-      exp = exp * 10 + (*p++ - '0');
-    } while (p != end);
-    if (sign == '-') exp = -exp;
-    int fraction_size = 0;
-    if (exp_pos != begin + 1) {
-      // Remove trailing zeros.
-      auto fraction_end = exp_pos - 1;
-      while (*fraction_end == '0') --fraction_end;
-      // Move the fractional part left to get rid of the decimal point.
-      fraction_size = static_cast<int>(fraction_end - begin - 1);
-      std::memmove(begin + 1, begin + 2, to_unsigned(fraction_size));
-    }
-    buf.try_resize(to_unsigned(fraction_size) + offset + 1);
-    return exp - fraction_size;
+    buf.try_reserve(size + offset + 1);  // Add 1 for the terminating '\0'.
   }
 }