In this study we examined the effect of: observer presence; observer speed; and the impact that multiple surveys had on the number of counted fish in a visual census survey in a typical Mediterranean rocky habitat. SCUBA observer presence had no influence on the number of fish per survey. An improvement in the methodology to estimate the number of fish was demonstrated by using two different speeds: slow speed for benthic species and higher speed for epibenthic ones. No differences in the number of fish per survey were found among successive surveys along the same transects, even with the shortest time interval used: 5 min. We also examined the ability of observers to estimate fish length during visual census by using wooden fish models. Significant differences in size estimations were found using two different estimation methods (i.e. size-classes estimates and continuous measures). Estimating fish length to a size class was more accurate in estimating size, while transformed size to calculate biomass from length estimates was more accurate using continuous measures.