The stringent quality requirement of petroleum products in a highly competitive market makes on-line monitoring and control of product properties essential. But unfortunately few on-line hardware sensors are available and these are also difficult to maintain. It is, therefore, necessary to develop 'software sensors' to predict the quality using other easily measurable secondary variables. This study presents an algorithm that uses the crude true boiling point (TBP) curve and other routinely measured flow rates, temperatures and pressures in the crude distillation unit (CDU) to predict the product properties. The measured top plate, side-stripper draw plates and flash zone temperatures are corrected for hydrocarbon partial pressures to obtain equilibrium flash vaporization (EFV) temperatures. These product EFVs are converted to product TBPs and are superimposed on the crude TBP curve. An assumption, that the initial boiling point (IBP) of the next heavier product lies vertically below the final boiling point (FBP) of the product under consideration and the two points are equidistant from the crude TBP curve, allows estimation of the IBP and FBP temperatures of all the distillate products. A straight line approximation of the product TBP curve is used to obtain intermediate temperatures. These TBP temperatures are converted to product ASTM (American Society for Testing Materials) temperatures which are correlated with the desired product properties. Several properties have been predicted using the above procedure. These include densities of all the CDU products, Flash Points for all the side-stream products, Reid Vapor Pressure (RVP) for the distillate, Freeze Point for kerosene, Pour Point and the recovery for the gas oils etc. It is possible to predict these properties repeatedly every minute as long as steady state conditions prevail in the CDU. The algorithm has been applied off-line with the available on-line data from two different operating refineries. A satisfactory match between the predicted and the measured properties validated the developed soft sensors. However, extensive testing is recommended before the implementation of these soft sensors on the actual process. (C) 2003 Elsevier Ltd. All rights reserved.