Coffea canephora Pierre breeding in Côte d'Ivoire is based on indirect reciprocal recurrent selection (RRSi) using the Congolese and Guinean populations. In practice, the genotypes of each population (78 Congolese genotypes and 100 Guinean genotypes) are tested with several testers from the reciprocal population: two Guinean and three Congolese testers. These testers are heterozygous as C. canephora is a strictly outcrossing species. After a first cycle of RRSi, we judged the efficiency of these testers for yield, susceptibility to coffee leaf rust (SCLR) and hundred bean weight (HBW). All the testers discriminated the tested genotypes highly significantly (P < 0.01), even though differences were found in the degrees of discrimination. For yield and SCLR, no interaction between tested genotypes and testers was observed and correlations between test values obtained from different testers were significant or highly significant. Furthermore, test values of tested genotypes could be used to predict the yield and SCLR of between-population (BP) hybrids. Prediction was always more accurate with the best tester than with the mean of several testers. As opposed to yield and SCLR, HBW showed significant interactions between tested genotypes and testers and HBW of hybrids could not be predicted from the test value of the genotypes. We conclude that only one reciprocal tester can be used in the next cycles of RRSi applied to C. canephora. This will allow either a reduced cost of selection or increased selection intensity.