The present study is based on data recorded from fish of the third generation of the GIFT project (Genetic Improvement of Farmed Tilapias). The objective of the study was to compose a synthetic base population of Nile tilapia for further selective breeding, and to estimate phenotypic and genetic parameters in that population. The base population was formed by choosing parent stock among the best performing animals within the best strain combinations resulting from a full diallel cross design (8 × 8) involving four African wild strains and four Asian farmed strains (64 strain combination cells altogether). The grandparent ancestors of the base population were mainly representing the three strains originating from Egypt, Kenya and Thailand (20.2, 27.3 and 19.6%, respectively). The proportion of ancestors from the Senegal strain was medium (12.2%) whereas the proportion of ancestors from Ghana, Israel, Singapore and Taiwan strains was low (3 to 8%) and mainly due to a restriction that all parent strains should be represented in the synthetic base. Statistical analyses using a mixed animal model were carried out on records from 13,570 individually tagged and surviving progeny of 50 sires and 123 dams. The growth performance of the fish was tested in seven diverse environments. The test environments covered a wide range of tilapia farming systems, namely, earthen ponds fertilized with inorganic fertilizer and organic manure or on-farm agricultural residues, cage culture, and test stations located in different agro-climatic regions. The model fitted included the fixed effect of test environment and sex and the linear and quadratic covariate of age at stocking. The random terms were the additive genetic effect of individual fish and the additional effects common to full-sib groups. The estimates of heritability for body weight at harvest in different test environments ranged from moderate to high (0.12 to 0.56) and the estimate across test environments was 0.15. The variances due to other effects common to full-sib families were significant, accounting for 8 to 16% of total variation in body weight. The genetic correlations among body weights recorded in similar environments were high (mostly > 0.80, i.e. among the earthen ponds in experimental locations). By contrast, the genetic correlations between harvest weight in earthen pond and in cage environments were more variable (0.36–0.82). Overall, the results suggested that the genotype by environment interactions were of limited importance, at least for the target pond farming systems in the Philippines. It was concluded that as the additive genetic variation in the base population was large, the harvest weight of the GIFT fish could be effectively improved by selective breeding, and that there was no immediate need to develop separate strains for different environments.