The data underlying scientific papers should be accessible to researchers both now and in the future, but how best can we ensure that these data are available? Here we examine the effectiveness of four approaches to data archiving: no stated archiving policy, recommending (but not requiring) archiving, and two versions of mandating data deposition at acceptance. We control for differences between data types by trying to obtain data from papers that use a single, widespread population genetic analysis, STRUCTURE. At one extreme, we found that mandated data archiving policies that require the inclusion of a data availability statement in the manuscript improve the odds of finding the data online almost 1000-fold compared to having no policy. However, archiving rates at journals with less stringent policies were only very slightly higher than those with no policy at all. We also assessed the effectiveness of asking for data directly from authors and obtained over half of the requested datasets, albeit with ∼8 d delay and some disagreement with authors. Given the long-term benefits of data accessibility to the academic community, we believe that journal-based mandatory data archiving policies and mandatory data availability statements should be more widely adopted.