"We propose newly developed unsmoothing techniques which are based on a regime-switching Threshold Autoregressive (TAR) model. We first examine analytically conventional unsmoothing techniques which model the true returns by a linear Autoregressive (AR) process ñ and show that when true returns follow a TAR process, the conventional technique is misspecified, and hence still underestimates true variance. Our approach also addresses identification problems that may occur in the conventional method. The threshold/regime technique can be applied both to underlying return processes and to smoothing behaviour. We test these individually and in combination for UK commercial real estate returns, using a variety of exogenous variables. Two variables, FT all share returns and LIBOR interest rates, provide results that improve on the conventional single regime model. The results are intuitively plausible, capturing return behaviour and suggest that in extreme ""crisis"" regimes, returns fall explosively and smoothing effects intensify, by comparison to more benign regimes. These results support parallel work on asymmetries in real estate behaviour and shed important light on the risks of real estate and its diversification potential."