A compositional model for predicting the viscosity of petroleum fluids as a function of temperature and pressure has been developed. It is based on a mixing rule that takes into account the compositional information, and on the model of Kanti et al. [1] for the dependence on pressure and temperature. The approach of Grunberg and Nissan [2] is used in order to take into account the fluids composition. Four different fractions are used : Gas, C-6-C-20, C20+ (saturates + aromatics + resins), and C20+ (asphaltenes). This model is designed so as to be valid in a large compositional range, especially for liquids containing large amounts of asphaltenes. The model of Kanti uses a single viscosity value for predicting fluid viscosity at any pressure and temperature. Its universal parameters have been fitted on a large database ranging up to 100 MPa in pressure and up to 120 degrees C in temperature. The validation of this new model on 76, so far, unpublished experimental data points of crude oils, covering a dynamic viscosity range from 2 to 23,298 mPa s, led to an absolute mean deviation of 24%,