The use of float aside, I think there is a problem with your statement.

If you are converting numeric(5,1) to float then you will get either an exact value or an approximation with a lot of decimal digits. The problem you stated with 100.1 is not a problem as 100.1 in a numeric(5,1) will be converted to 100.09999999999999 in a float and then 100.1 when converted back to numeric(5,1). If you convert 100.09 to numeric(5,1) you will get 100.1 due to rounding. Any value stored in a numeric(5,1) converted to float and back again will always give you the original value, the only time rounding can take place is when the value stored is not originated from the datatype used.

As for the overflow, I agree with Kenneth, you must have a value larger than numeric(5,1).

Kenneth, as for float and real being synonyms, that is not quite true, real is a synonym of float(24) and is only 4 bytes and you cannot specify the mantissa whereas with real you can and it can be upto 8 bytes.