Recovering market beliefs, time and risk preferences from asset prices requires specifying a subjective state space for the pricing model. Recovery results are unique to each pecification but differ cross specifications and can only be mutually consistent when each specification extracts identical price information. Finer specifications do not improve the recovery consistency because they require more price information, and hence, are also more sensitive to information distortion. Using option price data, analytical methods and neural network analysis, we show that recovery inconsistencies are common and economically significant. Our findings suggest that a fully consistent recovery framework remains difficult to achieve.