The purpose of this study was to examine how market participants form their inflation expectations in the Iranian economy over the period 1959-2003. Inflation expectations are very unstable in Iran's economy because the Central Bank is unable to adhere to an inflation target in practice. Thus, inflation expectations are not well-anchored and any oil price increase, which seems apparently to be a favorable shock, results in money creation, fueled by government spending out of oil revenues, and inflation and causes private agents to raise inflation expectations. This in turn will increase inflation. As a result, poor anchored inflation expectations make price stability much more difficult to achieve in the long run and decrease the Central Bank's ability to stabilize output and employment in the short run. This study compared two approaches to modeling inflation expectations: simple forecast and a multi-equation model. The results of simple statistical predictors revealed that the Neural Network model yields better estimates of inflationary expectations than do parametric autoregressive moving average (ARMA) and linear models. The agents were assumed to use a parametric autoregressive moving average (ARMA) model, proposed by Feige and Pearce (1976), or nonparametric models to form their expectations. Comparing to the nonparametric alternatives, the results of Wilcoxon tests demonstrated that the forecasting performance of Projection-Pursuit Regression and Additive models appear to differ from the Neural Network model, implying that the Neural Network model can significantly outperform Projection-Pursuit Regression model and it has a better performance than Additive model, but not by much. However, there was no possibility that the Neural Network model can outperform the Multiple Adaptive Regression Splines model. The results of estimated multi equation model indicated the expectation hypothesis is accepted for the models under backward-looking expectations and learning approach. Among near rational expectation schemes and the learning approach, the learning model was better suited for modeling inflation expectations than other alternative methods. Therefore, the Central Bank should be more aggressive towards inflation. Furthermore, as any decrease in inflation is highly desirable and is one of the main macroeconomic goals, solidly anchored inflation expectations are suggested. To do so we need to keep monetary policy tight for a considerable period.