diff --git a/sqlx-core/src/mysql/protocol/row.rs b/sqlx-core/src/mysql/protocol/row.rs
index e85464d5..d266f5c6 100644
--- a/sqlx-core/src/mysql/protocol/row.rs
+++ b/sqlx-core/src/mysql/protocol/row.rs
@@ -104,8 +104,8 @@ impl Row {
                 let size = match columns[column_idx] {
                     TypeId::TINY_INT => 1,
                     TypeId::SMALL_INT => 2,
-                    TypeId::INT => 4,
-                    TypeId::BIG_INT => 8,
+                    TypeId::INT | TypeId::FLOAT => 4,
+                    TypeId::BIG_INT | TypeId::DOUBLE => 8,
 
                     TypeId::DATE => 5,
                     TypeId::TIME => 1 + buffer[index] as usize,
diff --git a/sqlx-core/src/mysql/types/float.rs b/sqlx-core/src/mysql/types/float.rs
index 831e7ad5..f708424f 100644
--- a/sqlx-core/src/mysql/types/float.rs
+++ b/sqlx-core/src/mysql/types/float.rs
@@ -5,6 +5,19 @@ use crate::mysql::types::MySqlTypeInfo;
 use crate::mysql::MySql;
 use crate::types::HasSqlType;
 
+/// The equivalent MySQL type for `f32` is `FLOAT`.
+///
+/// ### Note
+/// While we added support for `f32` as `FLOAT` for completeness, we don't recommend using
+/// it for any real-life applications as it cannot precisely represent some fractional values,
+/// and may be implicitly widened to `DOUBLE` in some cases, resulting in a slightly different
+/// value:
+///
+/// ```rust
+/// // Widening changes the equivalent decimal value, these two expressions are not equal
+/// // (This is expected behavior for floating points and happens both in Rust and in MySQL)
+/// assert_ne!(10.2f32 as f64, 10.2f64);
+/// ```
 impl HasSqlType<f32> for MySql {
     fn type_info() -> MySqlTypeInfo {
         MySqlTypeInfo::new(TypeId::FLOAT)
@@ -23,6 +36,10 @@ impl Decode<MySql> for f32 {
     }
 }
 
+/// The equivalent MySQL type for `f64` is `DOUBLE`.
+///
+/// Note that `DOUBLE` is a floating-point type and cannot represent some fractional values
+/// exactly.
 impl HasSqlType<f64> for MySql {
     fn type_info() -> MySqlTypeInfo {
         MySqlTypeInfo::new(TypeId::DOUBLE)
diff --git a/tests/mysql-types.rs b/tests/mysql-types.rs
index 4fa15a50..01d9192a 100644
--- a/tests/mysql-types.rs
+++ b/tests/mysql-types.rs
@@ -18,11 +18,11 @@ macro_rules! test {
                     .fetch_one(&mut conn)
                     .await?;
 
-                assert_eq!(row.get::<i32, _>(0), 1);
-
                 let value = row.get::<$ty, _>("_1");
 
-                assert!($value == value);
+                assert_eq!(row.get::<i32, _>(0), 1, "value returned from server: {:?}", value);
+
+                assert_eq!($value, value);
             )+
 
             Ok(())
@@ -44,6 +44,9 @@ test!(mysql_long: i32: "2141512" == 2141512_i32);
 test!(mysql_longlong_unsigned: u64: "2141512" == 2141512_u64);
 test!(mysql_longlong: i64: "2141512" == 2141512_i64);
 
+// `DOUBLE` can be compared with decimal literals just fine but the same can't be said for `FLOAT`
+test!(mysql_double: f64: "3.14159265" == 3.14159265f64);
+
 test!(mysql_string: String: "'helloworld'" == "helloworld");
 
 #[cfg_attr(feature = "runtime-async-std", async_std::test)]
@@ -69,3 +72,23 @@ async fn mysql_bytes() -> anyhow::Result<()> {
 
     Ok(())
 }
+
+#[cfg_attr(feature = "runtime-async-std", async_std::test)]
+#[cfg_attr(feature = "runtime-tokio", tokio::test)]
+async fn mysql_float() -> anyhow::Result<()> {
+    let mut conn = connect().await?;
+
+    let value = 10.2f32;
+    let row = sqlx::query("SELECT ? as _1")
+        .bind(value)
+        .fetch_one(&mut conn)
+        .await?;
+
+    // comparison between FLOAT and literal doesn't work as expected
+    // we get implicit widening to DOUBLE which gives a slightly different value
+    // however, round-trip does work as expected
+    let ret = row.get::<f32, _>("_1");
+    assert_eq!(value, ret);
+
+    Ok(())
+}