Despite its wide use in terrestrial ecology, distance sampling is as yet rarely used in underwater visual censuses. The present study attempts to compare density estimators based on distance sampling and on strip transects. Three stations with increasing densities of Chaetodontidae and Pomacanthidae were sampled twice by two divers of unequal experience, using two different transect types. A total of 96 transects and 2970 records of Chaetodontidae and Pomacanthidae were analysed. Nine estimators based on distance sampling were calculated and only the best fit (DT estimator) was kept for comparison with other estimators. These were either based on the average distance of the fish to the transect (AD estimator), or a 3-m- or 5-m-wide strip transect estimator (FW3 and FW5, respectively). There were no significant differences between the means found by DT, AD and FW3. Lower density estimates were given by FW5 in all cases. FW3 and FW5 did not detect several significant differences between stations which were otherwise detected by DT or AD. The number of transects needed to detect a significant difference between stations was four to ten times higher with FW3 or FW5 than with DT or AD. Diver experience was found to be a significant factor in density estimates. However, this factor was less important than the choice of the density estimator. Transect type or the day of sampling had no consequence for the estimates. The distance distributions of fish were divided into three different patterns which may be explained by a combination of detectability function and a behavioural component.