It is well-established that certain occupations that have frequent contact with patients (e.g. health care workers) or the public (e.g. bus drivers), may have increased risks of COVID-19. However, estimating these risks involves a number of important methodological problems. Not everyone who has a SARS-CoV-2 infection gets symptoms; not everyone with symptoms gets tested; not all tests yield valid results. Therefore, even just estimating the incidence or prevalence of COVID-19 in particular occupational groups is difficult. Moreover, these selection pressures may be different for different occupational groups (e.g. health care workers may be more likely to be tested than some other occupations). In addition, unlike most other occupational exposures, you can get infected at work, or get the same infection at home, potentially yielding quite severe confounding by lifestyle and living conditions. These methodological issues are not insurmountable, but require careful study design and data analysis. In particular: (i) when not everyone is being tested, then comparisons between and within occupations can be biased, but this bias can be minimised using the test-negative design; and (ii) comparisons between occupations require careful and rigorous adjustment for work-related ‘living conditions’, i.e. the fact that workers in insecure low-paid jobs may also have a high risk of contracting SARS-CoV-2 infections outside of the workplace.