It is often claimed that up to 90% of low back pain (LBP) episodes resolve spontaneously within 1 month. However, the literature in this area is confusing due to considerable variations regarding the exact definitions of LBP as well as recovery. Therefore, the claim--attractive as it might be to some--may not reflect reality. In order to investigate the long-term course of incident and prevalent cases of LBP, a systematic and critical literature review was undertaken. A comprehensive search of the topic was carried out utilizing both Medline and EMBASE databases. The Cochrane Library and the Danish Article Base were also screened. Journal articles following the course of LBP without any known intervention were included, regardless of study type. However, the population had to be representative of the general patient population and a follow-up of at least 12 months was a requirement. Data were extracted independently by two reviewers using a standard check list. The included articles were also independently assessed for quality by the same two reviewers before they were studied in relation to the course of LBP using various definitions of recovery. Thirty-six articles were included. The results of the review showed that the reported proportion of patients who still experienced pain after 12 months was 62% on average (range 42-75%), the percentage of patients sick-listed 6 months after inclusion into the study was 16% (range 3-40%), the percentage who experienced relapses of pain was 60% (range 44-78%), and the percentage who had relapses of work absence was 33% (range 26-37%). The mean reported prevalence of LBP in cases with previous episodes was 56% (range 14-93%), which compared with 22% (range 7-39%) for those without a prior history of LBP. The risk of LBP was consistently about twice as high for those with a history of LBP. The results of the review show that, despite the methodological variations and the lack of comparable definitions, the overall picture is that LBP does not resolve itself when ignored. Future research should include subgroup analyses and strive for a consensus regarding the precise definitions of LBP.