diff --git a/dbms/src/Functions/FunctionsTiDBConversion.h b/dbms/src/Functions/FunctionsTiDBConversion.h
index af67521540e..abb88c2f638 100644
--- a/dbms/src/Functions/FunctionsTiDBConversion.h
+++ b/dbms/src/Functions/FunctionsTiDBConversion.h
@@ -1192,8 +1192,20 @@ struct TiDBConvertToDecimal
             /// cast enum/int/real as decimal
             const typename ColumnVector<FromFieldType>::Container & vec_from = col_from->getData();
 
-            for (size_t i = 0; i < size; ++i)
-                vec_to[i] = toTiDBDecimal<FromFieldType, ToFieldType>(vec_from[i], prec, scale, context);
+            if constexpr (std::is_integral_v<FromFieldType>)
+            {
+                /// cast enum/int as decimal
+                for (size_t i = 0; i < size; ++i)
+                    vec_to[i] = toTiDBDecimal<FromFieldType, ToFieldType>(vec_from[i], prec, scale, context);
+            }
+            else
+            {
+                /// cast real as decimal
+                static_assert(std::is_floating_point_v<FromFieldType>);
+                for (size_t i = 0; i < size; ++i)
+                    // Always use Float64 to avoid overflow for vec_from[i] * 10^scale.
+                    vec_to[i] = toTiDBDecimal<Float64, ToFieldType>(static_cast<Float64>(vec_from[i]), prec, scale, context);
+            }
         }
         else
         {
diff --git a/tests/fullstack-test/expr/cast_float_as_decimal.test b/tests/fullstack-test/expr/cast_float_as_decimal.test
new file mode 100644
index 00000000000..8948e029a7d
--- /dev/null
+++ b/tests/fullstack-test/expr/cast_float_as_decimal.test
@@ -0,0 +1,8 @@
+mysql> drop table if exists test.t1;
+mysql> create table test.t1(c1 float);
+mysql> insert into test.t1 values(3.40282e+37);
+mysql> alter table test.t1 set tiflash replica 1;
+func> wait_table test t1
+mysql> set @@tidb_isolation_read_engines='tiflash'; set @@tidb_enforce_mpp = 1; select cast(c1 as decimal(50, 2)) from test.t1;
+cast(c1 as decimal(50, 2))
+34028199169636079590747176440761942016.00