To describe the accuracy of clinical features and tests used to screen for malignancy in patients with low back pain. A systematic review was performed on all available records on MEDLINE, EMBASE, and CINAHL electronic databases. Studies were considered eligible if they investigated a cohort of low back pain patients, used an appropriate reference standard, and reported sufficient data on the diagnostic accuracy of tests. Two authors independently assessed methodological quality and extracted data to calculate positive (LR+) and negative (LR-) likelihood ratios. Six studies evaluating 22 different clinical features and tests were identified. The prevalence of malignancy ranged from 0.1 to 3.5%. A previous history of cancer (LR+ = 23.7), elevated ESR (LR+ = 18.0), reduced hematocrit (LR+ = 18.2), and overall clinician judgement (LR+ = 12.1) increased the probability of malignancy when present. A combination of age > or =50 years, a previous history of cancer, unexplained weight loss, and failure to improve after 1 month had a reported sensitivity of 100%. Overall, there was poor reporting of methodological quality items, and very few studies were performed in community primary care settings. Malignancy is rare as a cause of low back pain. The most useful features and tests are a previous history of cancer, elevated ESR, reduced hematocrit, and clinician judgement.