In this issue, you will find a paper by Hartvigsen et al., entitled ‘Patterns of musculoskeletal pain in the population. A latent class analysis using a nationally representative interviewer-based survey of 4817 Danes’. This paper takes a new approach to studying multi-site pain by asking people to select their most troublesome pain site and then investigating patterns of co-occurring pain for people with each primary pain site. Over the past few years, there has been increased recognition that while some people experience pain as an isolated condition at a single site, many people experience pain in multiple sites, and that the presence of this multi-site pain is associated with more severe impact on the person and their lives. How we deal with this, both in terms of research and clinically, is still a topic for discussion. A previous approach to analysing multi-site pain has been to sum the number of pain sites each person reports, thereby regarding bodily pain as a continuous variable from no pain sites, one pain site, two pain sites, etc. This is a simple but effective method, which has shown that the impact of pain increases with increasing number of pain sites (Kamaleri et al., 2008a,b). One conclusion from this is that it is the number of sites of pain, rather than the pain location, that is the important factor. However, the paper in this issue (Hartvigsen et al., 2013) might lead us to question this. The researchers asked people with pain to identify their most troublesome pain site – the primary pain site – and then described the co-occurrence of pains (or absence of other pains) for people with pain at each primary pain site. Their findings show some interesting patterns, and indicate that the ‘other’ pain sites that people report may not be randomly distributed, but show different patterns according to the primary pain site identified. For example, spinal pain clustered together, e.g., people with primary pain in the lower back commonly also reported pain in the neck or upper back. There is still work to do to find out whether these clusters of pain sites are generalizable to other populations and settings. Schmidt and Baumeister (2007) applied a similar technique, also using latent class analysis to identify clusters of people with different patterns of co-occurring (or single site) pain. Those researchers found quite different results, e.g., back pain clustered with lower extremity pain rather than with other spinal pain. The difference might be explained by different methods – they did not stratify analyses by primary pain site – but only further work will tell us if that is the case. But, no matter how we analyse or interpret patterns of single site or multi-site pain, many questions still remain about how and when the pain develops in the first place. Epidemiological studies indicate that musculoskeletal pain, including pain at multiple sites, is present throughout the life course. Further work, ideally using data from long-term prospective studies, would help us towards better understanding of the development of the long-term pain experience. For example, does pain start at one site and spread to other (perhaps adjacent) sites over time, akin to the amplification model of back pain suggested by Raspe et al. (2003)? This would likely be accompanied by deteriorating psychological status and worsening general health, a pattern that can be seen with increasing number of pain sites. Or are the clusters fixed or stable over time, as might be indicated by the findings of Kamaleri et al. (2009), who found that reporting of number of pain sites over a 14-year period was relatively stable? If it is the latter, do people suddenly develop multi-site pain with its accompanying consequences? Answering questions like this might improve our understanding of the causes of pain over the long-term. Cross-sectional or short-term studies are not going to be much of help due to difficulties in disentangling the time sequence between pain and potentially causal factors, which both may be long-term, fluctuating and difficult to capture with one-off measurements. We need more analyses of the existing longitudinal studies and birth cohorts, and specifically designed long-term cohort studies, with measurements at multiple time points, to begin to get to grips with explaining the development of and changes in patterns and clusters of musculoskeletal pain over time.