Crop variety mixtures can provide many benefits, including pathogen suppression and increased yield and yield stability. However, these benefits do not necessarily occur in all mixtures, and the benefits of diversity may be compromised by disadvantages due to increased crop heterogeneity. In-field development of mixtures by assembling many combinations of crop genotypes without prior expectation about which genotypes need to be combined to produce well-performing mixtures results in prohibitively large designs. Therefore, effective tools are required to narrow down the number of promising variety mixtures, and to then identify in experiments which of these deliver the highest benefits. Here, we first review current knowledge about the mechanisms underlying effects in ecological diversity experiments and in current agricultural applications. We then discuss some of the principal difficulties arising in the application of this knowledge to develop good variety mixtures. We also discuss non-conventional approaches to solve some of these issues. In particular, we highlight the potential and limitations of trait-based methods to determine good variety mixing partners, and argue that nontraditional traits and trait-derived metrics may be needed for the trait-based approach to deliver its full potential. Specifically, we argue that good mixing partners can be identified using modern genetic and genomic approaches. Alternatively, good mixtures may be obtained by combining varieties that respond differently to environmental variation; such varieties could easily be identified in standard variety testing trials. Preliminary analyses show that niche differences underlying the different environmental responses can indicate functional complementarity and promote mixture yield and yield stability.