Periphyton communities are often the dominant primary producers and an important energy source to higher trophic levels in rivers and streams. Empirical models of periphyton biomass (chlorophyll a ) that have high predictive power are generally lacking. The goal of this research was to assess and improve the predictability of periphyton in rivers. A historical river monitoring data set from Alberta showed that, in general, land use in the drainage basin was a good surrogate for instream nutrient concentrations in regression models of periphyton biomass. Land use explained up to 34% of the variability in chlorophyll a whereas models based on instream nutrients explained up to 24% of the variability in chlorophyll a . A field study showed that bacterial abundance in periphyton explained an additional 26 to 29% of residual variance of chlorophyll a , after taking nutrients into account. The relationships between algal and bacterial abundance and production estimates were positive, suggesting bacteria and algae coexist in a mutually dependent association. The sampling design for bacteria in the field study was based on the relationships between sample means and variances of published bacterial abundance and production data. The number of replicates needed to sample periphytic bacterial abundance and production was determined from these relationships. A meta-analysis of published periphyton regression models was used to evaluate model predictive power. Once corrected for the number of observations, terms, and sampling replicates in the models, predictive power of periphyton models has not improved over the last 30 years. Geographic extent of the study area and the type of predictor variables used also had almost no effect on predictive power. The theoretical limit of model precision has approximately been reached for models predicting temporally averaged periphyton biomass. In contrast, residual variances of models predicting instantaneous mean chlorophyll a were, on average, 4.5 times higher than theoretical pure error. Precision of temporal mean models will only be improved by obtaining more precise estimates of mean chlorophyll a . Models that predict instantaneous chlorophyll a may be made more precise by including variables that better reflect the recent site history.