Exposure-response relationships (ERRs) are of interest in many areas of epidemiology. In occupational and environmental epidemiology in particular, the nature of such associations may have practical implications for policy and the setting of standards. We use a study undertaken to estimate the association between neurobehavioral test scores and manganese exposure to illustrate that tests for linear trend may be significant in the presence of highly non-linear ERRs. We illustrate this point further with simulated data where the form of the ERR is known. We provide guidelines for exploring the nature of ERRs, in the absence of a priori knowledge of the response of the outcome of interest to exposure.