As macromolecular crystal structures are determined and refined in an increasingly automated fashion, careful assessment of the reliability and quality of the resulting models becomes increasingly important. Here, we analyze various issues related to the reliability and quality of macromolecular crystal structures deposited between 1991 and 2000. We find that the average resolution at which these structures are determined is essentially constant. In line with this observation, the average quality as measured by Ramachandran analysis does not improve as a function of time. On the other hand, an observed decrease of the average discrepancy between free and conventional R values suggests that the fit of model and data is improving. Finally, we present a surprising correlation between the tendency of crystallographers to deposit their experimental data and the free R values of their models.