Reliable estimation of effective fishing effort, which is proportional to fishing mortality, can provide information critical to the assessment and management of fisheries resources. To estimate effective fishing effort, we need to understand fishing efficiency and factors that may influence it. In this study fishing efficiency was estimated for mackerel and scad purse seine fisheries using generalized linear model. This was done for different companies involved in the fisheries. Different choices of error structures were considered in the estimation and their impacts on the estimation were discussed. The negative binomial distribution, gamma distribution, and lognormal distribution were chosen as error distributions according to loglinear regression of variance versus logmean of CPUE (catch per unit effort). Zero CPUE values had great impacts on the assumed error structure and adding a constant (δ) to CPUE was needed for gamma distribution and lognormal distribution in maximum likelihood estimation. As δ increased, the contrast of estimated fishing efficiency was reduced greatly. Delta approaches were also chosen as an alternative way to deal with zero CPUE value in this study. Comparing the results of different models, we considered Deltanegative binomial and Deltagamma as most appropriate error distributions for this study. The results showed that the fishing efficiency differed greatly among fisheries companies and among different areas.