<p>Predicting phenotypes from genomes is a major goal in genetics, but for most complex phenotypes, predictions are largely inaccurate. Here, we propose a more achievable alternative: relative prediction of phenotypic differences. Even with incomplete genotype-to-phenotype mapping, we show that it is often straightforward to determine whether an individual’s phenotype exceeds a threshold (e.g., of disease risk) or which of two individuals has a greater phenotypic value. We evaluated prediction accuracy on tens of thousands of individuals from the same family, same population, or different species. We found that the direction of a phenotypic difference can often be identified with &#xa0;&gt;90% accuracy. This approach also helps overcome some limitations in transferring genetic association results across populations. Overall, our approach enables accurate predictions of key information on phenotypes — the direction of phenotypic difference — and suggests that more phenotypic information can be extracted from genomic data than previously appreciated.</p>