Hedonic office rent prediction models which are the most common method based on multiple regression are well established in the literature. A wide range of variables, categorised as econometric, architectural, spatial and tenure rights, are used in these models for various cities. In the light of previous studies, some difficulties can arise in gathering data and applying the hedonic theory. As the dependent variable, asking rent is preferred in some models while contract rent or effective rent are used in others. It is reported that the use of contract or effective rent instead of asking rent, can provide more accurate predictions. However, it is difficult to obtain sufficient contract data from real estate firms, due to confidentiality and competition. The major difficulty lies within the hedonic regression models is the multicollinearity problem that may exist between a large number of independent variables. The common solution may reduce number of variables by exclude some variables depending on significance level or using ìstepwiseî or ìbackwardî procedure in regression models. In this study, it is attempted to construct a rent prediction model for _stanbul office market. The rent prediction model is improved in two ways. First, some variables are eliminated by ìbackwardî procedure in standard regression model and a reduced model is constructed. Second, factor analysis is conducted to group related variables and then, factors are incorporated into the regression model. Besides, three different rental values; asking rent, gross and net contract rent are used as dependent variable in the prediction models. Finally, performance of prediction models are compared according to R-squared and t-statistics. Akaike Information Criteria and Schwarz Information Criteria are also employed to test the accuracy of proposed models.