Abstract The accuracy of portion size estimates is critical in dietary assessment. The objectives of this study were to determine the magnitude and direction of error in estimating food amounts using computer-based portion anchors; to compare errors between use of computer- and poster-displayed anchors; and to assess level of accuracy that can reasonably be expected for different food types. A total of 101 subjects estimated amounts of food items (five solid, four amorphous, three liquid) using computer-based portion anchors. Half of the subjects also estimated five of the foods using digital photographs printed onto a poster. Two modes of testing were conducted: real time, in which foods were displayed near the computer or poster, and short-term recall, in which food was consumed and then estimated. Overall error (mean± se ) in real time was 4.8%±1.8%; by food type, solid was 8.3%±2.3%, amorphous was −10%±2.7%, and liquid was 19%±5%. However, group mean values could mask true errors. This research showed that even within food type, variables including amount served and its closeness to anchor size affected accuracy of estimates. For each food, amount served correlated inversely with magnitude of error. The findings provide insight into sources of error in portion estimation, and factors to consider in setting reasonable levels of accuracy.