Abstarct: In this work, quantitative structure-property relation ship (QSPR) has been applied for modeling and predicting the retention index of a series of polycyclic aromatic hydro-carbons (PAHs). The Polycyclic Aromatic Hydrocarbons (PAHs) have been for a long time, a focus of greater attention by the scientific community due to their impact on public health and the environment. Usually, the PAHs are introduced into the environment as a result of anthropogenic activities, which have increased dramatically in the last 20 years. PAHs are a class of several hundred individual compounds containing at least two condensed ring. PAHs are product of incomplete combustion and the pyrolysis of fusil fuel and other organic material from natural and anthropogenic source. Chromatographic techniques such as gas chromatography (GS) and high performance liquid chromatography (HPLC) coupled with UV, FID and MS detectors are the most frequently used. Retention index is a parameter that used in chromatography for separation of hydrocarbon mixture and provides the elution order of each compound. This order due of interaction between the mobile and stationary phase. This interaction related to the molecular structure of both phases and the variations in RI result from the structural variation of the compounds. Quantitative structure-properties relationships (QSPR) have extensively been used to explain separation mechanisms and prediction of retention behavior in chromatography. In the present work, we developed a QSPR model using artificial neutral network (ANN) and multiple linear regressions (MLR) to predict the retention index of some PAHs compounds. The results of this study also demonstrate the underling factors governing on the retention index of test molecules. The experimental retention indexes of 83 polycyclic aromatic hydrocarbons were taken from the published experimental paper. All molecules were drawn into HyperChem and pre-optimized using semi-empirical AM1 method. The resulted geometry was transferred into software Dragon that can calculate the different descxriptors.. The result obtained reveals the superiority of ANN over the MLR model.