A predictive method, based on quantitative structure-activity relationship (QSAR) techniques, has been developed for liquid viscosities of organic compounds. On the basis of the set of data including viscosity and other 18 physicochemical properties of 116 organic compounds of diverse structure over a viscosity range of 0.197-19.9mPa . s, two basic models using both multiple linear regression and partial least squares regression have been developed and their prediction abilities compared. The results recommend the traditional multiple linear regression model. The basic model has been developed further to cover about 230 compounds. Required parameters for the predictive model can be readily available or calculated purely from structural information. The prediction result is critically compared with four existing approaches in versatility and reliability. This approach can be used for the reasonably accurate prediction of liquid viscosities for a wide variety of organic compounds based on chemical structure.