Accuracy and bias of visual estimates of the number, size structure and biomass of juveniles of a small coral reef fish, Salarias fasciatus, were examined on 10 artificial patch reefs at One Tree Reef, Great Barrier Reef. Visual estimates of the number and length of flsh were made by 3 observers for 3 different durat~ons (5, 10 and 15 min) at each reef. These estimates were compared to measurements obtained from subsequent destructive sampling of reefs (98.5 % of flsh collected from all reefs). An increase in census penod does not continue to improve the accuracy of the estimates of numbers of fish The accuracy of the 10 min census did not differ significantly from the 15 min census, although it was signif~cantly higher than the 5 min census. Also, observers had unique biases for specific size-classes when estimating the size structure of the population. Biomass was estimated from the visual censuses of one observer, using a length-weight relationship, and compared to a known bion~ass from subsequent destructive collections. The mean accuracy of the biomass estimate derived from visual estimates of lengths was 72.3 %. This level of accuracy was obtained when the length-weight relationship was good (r2 = 0.97) and the estimates of lengths of each size-class (mean accuracy = 86 %, pooled for observer, times and sites) and the estimate of numbers (mean accuracy = 91.690 for the 10 min census) both had high levels of accuracy. But such high accuracies may not b e expected for censuses of larger mobile species. The use of visual censuses to estimate the biomass of such fish populations are likely to b e subject to greater inaccuracy. Most of the error in the visual estimates of biornass was explained by the inaccuracies of estimating the length of fish visually. Categorising fish into 5 mm sizeclasses or using a length-weight relationship to convert length to biomass did not significantly affect the accuracy of the estimates.