Abstract It has become clear that many wellbore stability studies suffers from various inconsistencies. These lead to incorrect results, or results that may not be extrapolated to other well configurations. In this paper typical wellbore fracturing and collapse models are investigated with respect to the accuracies in the input data. It is shown that the cumulative uncertainty is considerable. To reduce this uncertainty, constraints on the in-situ stress states are invoked. The paper further advocates the need for model calibration. In addition to using leak-off, breakout and wellbore image data, methods such as establishing bounds on the in-situ data, data normalization and inversion methods are proposed to improve accuracy and to provide quality assurance. Examples are provided in the paper.