This paper used eight groups of numerical data, which were composited by some different length of digits, to demonstrate the existance of truncation error in the computer operation. The IBM 360 computer was used as the instrument. It provided 7 significant digits in the single-precision floating point mode, and 16 in double-precision flotaing point mode. The findings show when the length of digits was long, using definition formula in both single and double precision floating point mode would get more precise outputs than using the computation formula; and there was no truncation error, if the length of digits was short.