Sources of variability of krill mean length and fish bycatch number was investigated to explore proper allocation of scientific observers on Japanese krill fishing vessels using the data from 1995 to 2008. Based on the multistage sampling theory, we derived the equation of variance in four stage cluster sampling units (i.e., vessel, year, subarea, and haul), and examined changes in the coefficients of variation (CV) caused by the shifts in coverage at different sampling stages. Vessel coverage had the largest impacts, and year coverage had moderate impacts on CV for both mean krill length and fish bycatch. The impacts of subarea and haul coverage were different between krill length and fish bycatch. Fish bycatch required moderate coverage for haul but low improvement were made by increasing subarea coverage, whereas krill length allowed very low haul coverage but preferred higher subarea coverage. To determine the proper level of observer coverage, the purpose and cost of observer monitoring should be discussed.