Few replicable genetic associations for Major Depressive Disorder (MDD) have been identified. Recent studies of MDD have identified common risk variants by using a broader phenotype definition in very large samples, or by reducing phenotypic and ancestral heterogeneity. We sought to ascertain whether it is more informative to maximize the sample size using data from all available cases and controls, or to use a sex or recurrent stratified subset of affected individuals. To test this, we compared heritability estimates, genetic correlation with other traits, variance explained by MDD polygenic score, and variants identified by genome-wide meta-analysis for broad and narrow MDD classifications in two large British cohorts - Generation Scotland and UK Biobank. Genome-wide meta-analysis of MDD in males yielded one genome-wide significant locus on 3p22.3, with three genes in this region (CRTAP, GLB1, and TMPPE) demonstrating a significant association in gene-based tests. Meta-analyzed MDD, recurrent MDD and female MDD yielded equivalent heritability estimates, showed no detectable difference in association with polygenic scores, and were each genetically correlated with six health-correlated traits (neuroticism, depressive symptoms, subjective well-being, MDD, a cross-disorder phenotype and Bipolar Disorder). Whilst stratified GWAS analysis revealed a genome-wide significant locus for male MDD, the lack of independent replication, and the consistent pattern of results in other MDD classifications suggests that phenotypic stratification using recurrence or sex in currently available sample sizes is currently weakly justified. Based upon existing studies and our findings, the strategy of maximizing sample sizes is likely to provide the greater gain.