Estimation models are used to evaluate of variance components in studies of sugarcane families. This paper compared the MIVQUE (0) and REML random models and LSMEANS and PMF, fixed models. Concerning the ranks of families to TCH and BRIX, low changes were observed between the two random estimators and LSMEANS. To the PMF, which considers the standard deviations within families, there was exchange of ranks. The range of prediction between families with MIVQUE (0) approach were 2.98 and 32.72, respectively to Brix and TCH, showing the existence of variability between families. To the individual predictions (progenies), only the method MIVQUE (0) was used, due the sample size, 200 progenies per family showed an efficient prediction model. From 365 selected progenies, 305 are among the highest predicted value to TCH and to BRIX only 37, suggesting the need to adopt combined strategies for identifying elite clones. The estimation of the heritability mean between families for hypothesis of 80 progenies was 0.64 for TCH and 0.86 for Brix. Indicating efficiency in the selection between and within families. Stalk number in ratoon cane cycle was the main biometric feature associated and strongly correlated with TCH and biomass, 0.79 and 0.85 respectively.