The paper discusses the representation of expectation formation processes in system dynamics. After a brief overview of current behavioural research on expectation formation, it analyses the implicit assumptions that arise from a representation with exponential smoothing and the TREND function. It addresses the limitations of univariate autoregressive algorithms and illustrates their difficulties in representing the causal reasoning processes that may underlie expectation formation. It is argued that exponential smoothing and TREND actually neglect the importance of causal and systemic reasoning and thus are not in line with the paradigm of systems thinking. Finally, three alternative approaches to modelling expectation formation are outlined.