AbstractWe investigated the effect of observer error on four commonly used species diversity measures: species richness, Shannon–Weiner diversity, Shannon–Weiner evenness and Simpson’s index of diversity. We also evaluated how observer error affects inferences derived from multivariate analyses of species–abundance relationships as determined by non-metric multidimensional scaling (NMS) ordination. Grassland vegetation was sampled by three different botanists at two national park units in Missouri and Kansas, USA. The same plots were sampled by two of the botanists, who compiled lists of species composition and estimated foliar cover. Differences in the data records were then compared. Pseudoturnover (i.e. apparent turnover due to observer error) ranged from 17.1% to 22.1%, and differences in cover class estimation ranged from 21.5% to 30.5%. The percentage difference in species diversity measures between pairs of observers depended on how data were summarized, but were always <20%, and often <10%. Based on these results, species diversity metrics are affected to a relatively smaller extent by observer error than turnover indices. Turnover indices, however, contain more information because they track individual species, whereas species are interchangeable in most species diversity indices. Thus, less of the error is identified because of how species diversity indices are calculated. NMS ordinations revealed that while the characterizations of some plots by different observers were similar, differences between observers’ records for other plots resulted in greater separation in ordination space. Points representing one observer’s records were often shifted in ordination space in the same direction compared with the other observer.