In production forests in the moist tropics, trees are selected for felling or retention primarily by species and size. Tree regeneration requirements and forest stand responses to harvesting are often ignored, and consequently, the regeneration of the residual forest is not ensured. We developed and tested an alternative approach to tree selection, where seed trees were retained as a proportion of harvestable trees, with the proportion defined as a function of species’ ecological attributes and local abundance (100ha), in contrast to the conventional approach which retained 10% of harvestable trees, uniformly across commercial species at the compartmental scale (1000ha). The study was conducted in Democracia Project, a forest management operation in Amazonas, Brazil. The conventional approach failed to retain any seed trees at the 100ha block scale for 7 of 37 commercial species, whereas the alternative approach retained a minimum number of seed trees per 100ha block for all commercial species. The conventional approach resulted in the retention of relatively high proportions of potential seed trees for common species (e.g., 22% for Eperua oleifera and 36% for Maquira sclerophila) that are shade bearers and recruit readily at the site; alternately, for species with constraints to regeneration, it retained relatively low proportions (e.g., 2% for Dinizia excelsa and Hymenolobium nitidum). The alternative approach effectively retained lower proportions of common species (e.g., 10% for E. oleifera and 13% for M. sclerophila) and relatively high proportions of species with regeneration constraints (e.g., 20% for D. excelsa and 16% for H. nitidum). Our study demonstrates that it is feasible to implement at an operational scale, species-specific retention rules that take into account local abundance when inventory data are digitised and spatially explicit. Monitoring regeneration in the residual stands over time will provide the evidence to assess the ecological benefits of the adoption of our alternative approach.