Appraisal-based performance indices are commonly used when considering private investments, including real estate. An important issue with such indices is the smoothing bias which leads to high levels of autocorrelation and dampened volatility estimates. The ability to treat appraisal-based series for smoothing is crucial for agents such as investors who have to correctly assess investment risk. In this research, we propose to improve the commonly used reverse engineering desmoothing model, as well as its regime-switching extension, by incorporating a robust filter into the procedure. We show that in addition to properly treating for smoothing, our improved model prevents the producing of extreme observations often generated through the desmoothing process. Our model is able to generate desmoothed series whose characteristics are akin to those of transaction-based indices. We also compare our results with those produced by alternative desmoothing models as well as those obtained by relying on REIT indices. The empirical analyses are conducted using U.S. data for a period extending over more than 30 years.