A Bayesian hierarchical approach is presented for the estimation of length-weight relationships (LWR) in fishes. In particular, estimates are provided for the LWR parameters a and b in general as well as by body shape. These priors and existing LWR studies were used to derive species-specific LWR parameters. In the case of data-poor species, the analysis includes LWR studies of closely related species with the same body shape. This approach yielded LWR parameter estimates with measures of uncertainty for practically all known 32 000 species of fishes. Provided is a large LWR data set extracted from, the source code of the respective analyses, and ready-to-use tools for practitioners. This is presented as an example of a self-learning online database where the addition of new studies improves the species-specific parameter estimates, and where these parameter estimates inform the analysis of new data.