The search for accuracy on biomass estimation is growing significantly, however density measurements are still essential to reduce bias in biomass estimations. The objective of this research was to: (1) test longitudinal and radial variations in wood density of Distemonanthus benthamianus, Musanga cecropioides and Trema orientalis. (2) Fitting mixed models of secondary forests pioneer species. Data for density determinations and fitting allometric equations were obtained by destructive method, on a sample of 115 trees. Archimedes' principle applied to the biomass data yielded the average densities of the three respective species (0.726; 0.214 and 0.35 in g.cm-3). This variable associated with tree height, crown diameter and diameter at breast height were used to explain tree biomass through ten fitted mixed models. The model, Aboveground biomass = Exp (-0.85 + 2.19 x ln (DBH) + 1.1 x ln (φ)), with a low Akaike Information Criterion (AIC = 78.76), the high correlation coefficient (Adjusted.R2 = 96.4%), the low rate of residual standard error (RSE = 0.33) and the Relative Root Mean Square Error (RRMSE = 0.39), was selected as the best mixed model. The full model (Aboveground biomass = Exp (-0.84 + 0.63 x ln (DBH2 x H) + 0.85 x ln (φ) + 0.54 x ln (C)) under the validation criteria was found to be efficient (adjusted R2 = 0.96; RRMSE = 0.41; average error = 15,95). However, density variations must be considered to reduce bias in the estimation. In addition, increased collection of large amounts of secondary forest data remains essential for fitting more robust mixed models.