Breeding cassava relies on a phenotypic recurrent selection that takes advantage of the vegetative propagation of this crop. Successive stages of selection (single row trial- SRT; preliminary yield trial – PYT; advanced yield trial – AYT; and uniform yield trials UYT), gradually reduce the number of genotypes as the plot size, number of replications and locations increase. An important feature of this scheme is that, because of the clonal, reproduction of cassava, the same identical genotypes are evaluated throughout these four successive stages of selection. For this study data, from 14 years (more than 30,000 data points) of evaluation in a sub-humid tropical environment was consolidated for a meta-analysis. Correlation coefficients for fresh root yield (FRY), dry matter content (DMC), harvest index (HIN) and plant type score (PTS) along the different stages of selection were estimated. DMC and PTS measured in different trials showed the highest correlation coefficients, indicating a relatively good repeatability. HIN had an intermediate repeatability, whereas FRY had the lowest value. The association between HIN and FRY was lower than expected, suggesting that HIN in early stages was not reliable as indirect selection for FRY in later stages. There was a consistent decrease in the average performance of clones grown in PYTs compared with the earlier evaluation of the same genotypes at SRTs. A feasible explanation for this trend is the impact of the environment on the physiological and nutritional status of the planting material and/or epigenetic effects. The usefulness of HIN is questioned. Measuring this variable takes considerable efforts at harvest time. DMC and FRY showed a weak positive association in SRT (r= 0.21) but a clearly negative one at UYT (r= -0.42). The change if the relationship between these variables is the result of selection. In later stages of selection, the plant is forced to maximize productivity on a dry weight basis either by maximizing FRY or DMC, but not both. Alternatively, the plant may achieve high dry root yield by simultaneously attaining acceptable (but not maximum) levels of FRY and DMC.