In about 5% of all cases LBP is associated with serious underlying pathology requiring diagnostic confirmation and directed treatment. Magnetic resonance imaging (MRI) is often used for this diagnostic purpose yet its role remains controversial. Consequently, this review aimed to summarize the available evidence on the diagnostic accuracy of MRI for identifying lumbar spinal pathology in adult low back pain (LPB) or sciatica patients. MEDLINE, EMBASE and CINAHL were searched (until December 2009) for observational studies assessing the diagnostic accuracy of MRI compared to a reference test for the identification of lumbar spinal pathology. Two reviewers independently selected studies for inclusion, extracted data and assessed methodological quality. Pooled summary estimates of sensitivity and specificity with 95% confidence intervals were calculated for homogenous subsets of studies. Eight studies were included in this review. Strata were defined for separate pathologies i.e. lumbar disc herniation (HNP) and spinal stenosis. Five studies comparing MRI to findings at the surgery for identifying HNP were included in a meta-analysis. Pooled analysis resulted in a summary estimate of sensitivity of 75% (95% CI 65-83%) and specificity of 77% (95% CI 61-88%). For spinal stenosis pooling was not possible. The results suggest that a considerable proportion of patients may be classified incorrectly by MRI for HNP and spinal stenosis. However, the evidence for the diagnostic accuracy of MRI found by this review is not conclusive, since the results could be distorted due to the limited number of studies and large heterogeneity.