If an observed length of fish is affected by error of mcasurement, the frequency distribution obtained would be deformed from the original one. The relation between observed length χ and exact length ξ may be assumed to be χ=ξ+δ, and also the distribution of error δ may be assumed to be normal with a constant variance of σE2. Then the expected frequency curve obtained from the observation would be given by the formula (4). Three types of distribution chosen here as examples are normal, exponential and rectangular. In the normal type, no bias appears except for an increase in Variance by σE2. On the contrary, in the exponential and rectangular types, considerable biases are introduced by the error. However, when compared with the sampling variance, the biases, are not so serious that the effect may be negligible for practical uses.