OBJECTIVES: Anchor-based methods are frequently used for determining the minimal important difference (MID) of scales employed to measure patient reported outcomes (PRO). The anchor may, e.g., consist of a global rating by the patient or the doctor or of a clinical measure closely related to the issue to be measured. The role of the psychometrical properties of the anchor has been rarely studied in this context. Aim of this contribution is to shed more light on the relationship between the reliability of the anchor and the estimated MID. METHODS: We performed a simulation study in which the reliability of the anchor used for MID estimation was varied systematically. Features of real-life data (e.g., skewed distribution, discreteness of PRO scale) and anchors were used to generate simulated PRO scales and anchors. MIDs were then estimated on the basis of the simulated data. RESULTS: Compared to the MID value obtained with an anchor with perfect reliability (r=1), a marked attenuation of the MID was observed when reducing the reliability of the anchor. Thus an anchor with reliability 0.7 gave rise to a 24-35% decrease of the MID estimate and an anchor with reliability 0.5 led to a 45-55% reduction. Based on the findings and on theoretical considerations we suggest a method for bias correction. CONCLUSIONS: When determining the MID of a PRO scale by an anchor-based method the reliability of the anchor plays a crucial role. Anchors with poor to moderate reliability may lead to considerable underestimation of the MID. Bias correction is possible provided the reliability of the anchor is known.