The purpose of this paper is to analytically and empirically show that the dimensions of SERVQUAL are not stable across service industries and even in the same service industry, because customers’ evaluations on service quality are heterogeneous. This paper presents a purification model that purifies the measured quality of service attributes through excluding the effects of service types and customer types on the quality levels of service attributes. It also derives SERVQUAL dimensions from the purified quality of service attributes via two alternative approaches (the factor-based approach and the cluster-based approach). Furthermore, based on the two alternative approaches, it analytically and empirically demonstrates that it is more logical to interpret that SERVQUAL dimensions may be heterogeneous across services, and that two grand dimensions (tangibles vs. non-tangibles) can be more stable than the five dimensions of SERVQUAL. This finding implies that there is no need to stick to the five dimensions of SERVQUAL, because they do not represent a customer’s psychometric dimensions of service quality.