The topic of customer loyalty in financial services is a perennial concern for both theoreticians and practitioners who share an interest in understanding its antecedents and in its prediction. The complex multidimensional causal structure of loyalty and its antecedents has attracted significant attention by researchers although there are still opportunities in developing causal models and in improving their predictive power. In this context, Structural Equation Models (SEM) would appear to be ideally suited, although the published track record of this approach is patchy. In this context, the distinction between formative and reflective approaches to SEM offers the prospect of improving the performance of SEM in modelling customer loyalty. This current empirical study compares the two approaches to modelling customer loyalty and draws some conclusions for SEM methodology. In this study, variety seeking, risk taking behaviour, and resistance to change are tested in explaining loyalty. The combined reflective and formative model was found superior over the purely reflective SEM approach.