An aerial survey of Pacific walruses (Odobenus rosmarus divergens) was evaluated to determine the reliability of estimates of population abundance. The probability of detecting groups of walruses on the pack ice remained uniform to at least 0.93 km from the flight line, whereas the probability of detection decreased significantly beyond 0.23 km for walruses in the water. Walruses were more abundant along the ice-edge zone between 162 and 165°W than in other areas of the Chukchi Sea during September 1975. Few walruses were observed in consolidated pack ice north of the ice-edge zone or in ice-free water to the south. More walrus groups and larger mean group size were observed on September 8 than on other days. We estimated abundance for each day and all days combined using methods based on sample area and numbers of strip samples. Estimates varied among days by over an order of magnitude; this variation is attributed to the combined effect of chance sampling of an aggregated population and variation in the fraction of walruses hauled out. The coefficient of variation of the estimates ranged between 0.25 and 0.99. This imprecision was due to the aggregated distribution of walruses and the large variation in group size. Using the survey data as a basis for stratification, we calculated that, due to the high variability within strata, a sample size of 40% of the total area or 56% of the total available strips would be required to obtain 95% confidence limits within 10% of the estimate of total abundance. Variation contributed by observer error in estimating group size also is relatively unimportant to the precision of abundance estimates. Studies of natural history, particularly those oriented toward activity and habitat selection, would help investigators estimate bias due to the variable fraction hauled out and design surveys based on meaningful strata. Estimates of total abundance based on limited survey efforts will provide information of little reliability.