Photographs are valuable aids in the estimation of food portion sizes and are easy to use in dietary surveys. As with other methods, employing photographs to estimate portion sizes consumed produces errors due to the method itself and errors from other sources such as poor recall of amounts actually eaten. This study was aimed at the first type of error and investigated errors due to the visual perception of food portions from food photographs in the absence of recall biases. Two hundred and seventy women were presented with various amounts of foods and asked to indicate the portion sizes using appropriate sets of photographs. The photographs showed three portion sizes (small, medium, large) for each of 45 foods commonly found in the French diet. The portions of real food estimated by subjects were prepared to the same weights as the portions photographed. Estimates of portion sizes were compared to the weighed amounts. They were found to be accurate within 25% in most cases. Except for three foods, errors on estimates were seen for one or more portions per food. Two patterns of errors were identified. For 22 foods, the small portion sizes were overestimated and the large portion sizes underestimated, but no error occurred with the medium portion. For 20 other foods, the three portion sizes were either all underestimated or all overestimated. The first pattern of error may be due to a general tendency to avoid extreme response categories, while the second pattern of error may be related to biased visual perception. In general, there was no obvious way to link the patterns of errors with the type or physical appearance of the foods. In conclusion, the photograph method is recommended for the assessment of portion sizes, provided that great care is taken to suggest the volume of the food portions in the photographs and in selecting a range of portions which encompasses the range of amounts of food actually consumed in the diet.