The accuracy of aerial estimates of avian aggregation sizes is variable across studies, and the relative importance of techniques and inter-observer error to this variation are poorly understood. Using a scaled physical model of a wading bird colony, we examined accuracy and variation in observer counts of simulated large numbers (200–6000) of densely nesting birds in vegetated situations. Observer estimates averaged 29% less than true numbers (under- and overestimates averaged together), and the mean absolute value of observer errors was 49% of true values. We found no effects of the size of the aggregation, the experience of the observer, the size of the previous aggregations surveyed by observers, the use of corrective lenses, or fatigue on degree of individual error. Over- and underestimates by individuals did not tend to cancel out in estimates by individuals of a “population” of colonies. Photographic counts of the same trials were significantly more accurate than observer estimates. We suggest that many studies using estimates of large numbers of birds may be confounded by similar errors. We urge that researchers use caution in interpreting the results of past surveys and develop ways to minimize, measure, and correct for visual estimation error within individuals and among observers.