National and international policies have encouraged the establishment of a representative network of marine protected areas (MPAs) in South Africa, with the aim of protecting marine biodiversity. The extent to which these marine and estuarine protected areas (EPAs) represent marine fish species and communities was assessed by comparing their species compositions with those of exploited areas, as sampled using four fishing techniques. Seven hundred fish species were sampled, representing one-third of South Africa's marine fishes. MPAs in coastal habitats scored c. 40% on the Bray-Curtis measure of similarity for species representativeness, but this score declined markedly for offshore ‘trawlable’ fishing grounds. The combined effects of sampling error, temporal variation and the effects of fishing on relative abundance suggest that 80% similarity would be the maximum achieveable. Forty-nine per cent of all fish species that were recorded were found in the 14 MPAs sampled. Redundancy in the MPA network was low, with fish species most commonly being represented in only one MPA or absent. There was greater redundancy in the 33 EPAs, with 40% of species being found in two or more EPAs, but many of these estuaries were adjacent to each other and embedded in large MPAs. Deep water fish communities (>80 m deep) and communities located on the west and south-east coasts of South Africa were most poorly represented by MPAs. Routine fishery surveys provide a robust and repeatable opportunity to assess species representativeness in MPAs, and the method used could form the basis of an operational definition of ‘representative’. In contrast to an assessment based on presence-absence data, this analysis of quantitative data presents a more pessimistic assessment of protection.