We analyzed the effect of outliers of the catch-per-unit effort on the catchability coefficient estimated by using a depletion model. When we used catch-per-unit effort in the Delury model, we observed a curve in the regression of depletion against time. When we then solved the model with a normal probability distribution, the catchability coefficient was poorly estimated. We improved the estimation of catchability using an algorithm that used a two-component-mixture probability distribution. The estimations for catchability (q) and recruitment (N ₀) were q = 0.41 × 10⁻³, N ₀ = 9.13 × 10⁶, and the estimated likelihood was 2.65 × 10⁴ using an algorithm of the normal probability distribution, whereas the estimations made using the algorithm of a two-component-mixture probability distribution were q = 0.23 × 10⁻³, N ₀ = 18.07 × 10⁶, and the estimated likelihood was 4.89 × 10⁶. The maximum likelihood estimated with the mixture-distribution algorithm was greater than the maximum likelihood estimated with the normal-distribution algorithm. We believe the two-component-mixture probability distribution fit the data better than the normal probability distribution. From this we determined the consequences on management when overestimations or underestimations of catchability are estimated.