In order to predict specific gravity and kinematic viscosity, distillation curves (used in the quality control of diesel) were associated with the partial least squares method (PLS). The regression coefficients of the algorithms were compared after selecting the appropriate number of latent variables for each prediction. The low values (0.82 kg m(-3) and 0.095 mm(2) s(-1)) obtained for RMSEP (Root Mean Squared Error Prediction) and the high correlation coefficients (0.9973 and 0.9943) between real and predicted values showed that PLS associated with distillation curves was efficient to determine specific gravity and kinematic viscosity. Models created with diesel samples of different composition were used for these predictions, thus reflecting the variety of fuel in the Brazilian market. In addition, the proposed method is low cost and easy to carry out, since it resorts to the results of a routine assay performed to assess the quality of highway diesel fuel. (C) 2012 Elsevier B.V. All rights reserved.