Shrinkage and coefficient of anisotropy are fundamental physical parameters in analysis of dimensional stability of timber structures members. In Brazil, properties of wood are obtained according to assumptions and calculation required by Annex B of ABNT NBR 7190: 1997 (Timber Structures Design). Among the physical properties, density appears as the easiest one to be experimentally obtained, defined by ratio between mass of the sample and its volume. To evaluate the possibility of estimating shrinkage (longitudinal, RL; radial, RR; tangential, RT; volumetric, RV; and coefficient of anisotropy, CA) of wood depending on density, this study aimed to test regression models (linear, exponential, logarithmic, geometric) using basic density (ρ bas ) and apparent density (ρ 12 ) as independent variables. For this purpose, five wood species belonging to conifers: Pinus sp., Pinus oocarpa; and dicotyledonous: Paricá (Schizolobium amazonicum), Jatobá (Hymenaea sp.) and Lyptus®, were employed. In all, 23 samples for Pinus sp.; 15 for Pinus oocarpa; 28 for Paricá; 16 for Lyptus®; and 11 for Jatobá were produced, totaling 651 determinations. Results of analysis of variance for regression models indicated insignificance of the adjustments obtained by adopting densities as estimators of shrinkage and coefficient of anisotropy for this set of wood species. Only models involving basic density and apparent density were significant, wherein the optimal settings were from linear case, exhibiting coefficient of determination (R 2 ) 91.72% (for softwoods) and 83.36% (for hardwoods).