Mendelian randomisation (MR) studies, which investigate causal effects of exposures on disease, might be biased by sample selection and misclassification if phenotypes are not measured universally with the same definition in all study populations or participants. For example, in MR analyses of effects of exposures on covid-19, studies might include individuals with specific characteristics (eg, high socioeconomic position) meaning they are more likely to be tested for SARS-CoV-2 infection or respond to study questionnaires collecting data on infection and disease (selection bias). Alternatively, studies might assume those who were not tested have not been infected by SARS-CoV-2 or had covid-19 and are included as control participants (misclassification bias). In this article, a set of analyses to investigate the presence of selection or misclassification bias in MR studies is proposed and the implications of these on results is considered. The effect of body mass index on covid-19 susceptibility and severity is used as an illustrative example.