Predicting fish community structure in streams is a challenge considering the strong dynamics of these environments. In this study, we tested whether using a fish dataset obtained in a previous time was relevant to predict fish community structure in a subsequent time. We also tested whether temporal beta diversity of fish communities was correlated with environmental variability, stream size and order. To test these hypotheses, we collected data on fish communities, environmental and spatial variables from 15 streams in the Rio das Mortes Basin (Mato Grosso State, Brazil) in two consecutive drought periods (in 2016 and 2017). The gradients in fish richness and abundance were correlated between years. The results of a variation partitioning analysis indicated that the fish community structure in 2016 was the main explanatory matrix of the fish community structure in 2017 (when compared to environmental and spatial variables). A variation partitioning analysis, based only on environmental and spatial variables, showed a much higher residual variation. We did not detect significant relationships between fish temporal beta diversity and our explanatory variables. Our results indicate that our predictive power may be substantially increased by using data on past communities as explanatory variables. This is a viable analytical strategy because long-term studies are becoming more frequent. Temporal autocorrelation analyses of community data can also be useful to evaluate priority effects. In addition, these analyses can help plan biomonitoring programs. The second part of the results indicates, however, that our ability to predict temporal beta diversity is still limited.