This paper presents a multivariate regression method for simultaneous detection of sugar (sucrose as a sugar equivalent) and ethanol concentrations in aqueous solutions via temperature-dependent ultrasonic velocity. Thus, several samples of different combined concentration values were exposed to a temperature spectrum ranging from 2 to 30°C to investigate the temperature dependence of ultrasonic velocity. Model calibration was performed in order to predict the concentrations of interest. With results of proceeded experiments, the equations for calculation of unknown concentrations were carried out using polynomial regression revealing two equations with functional dependence of concentrations on each other. Further, side effects or systematic errors are still included in this model. To avoid such problems as well as to increase the accuracy with respect to the absolute errors in determining unknown probes, multivariate regression methods such as partial least squares (PLS) were tested and compared to the results obtained by polynomial regression. The accuracy achieved with chemometric models on average was three times higher. In direct comparison, the values of the error for the prediction of sucrose concentration were on average around 0.4 g/100 g in the regression model with polynomial background (RMPA) and around 0.12 g/100 g in the PLS model, and for ethanol concentration 0.13 and 0.04 g/100 g, respectively. Furthermore, calculations of the concentrations are possible without knowing the concentrations of the other solute. Copyright © 2011 John Wiley & Sons, Ltd.