A pathologically involved margin in rectal cancer is defined as tumour within 1 mm of the surgical resection margin. There is no standard definition of a predicted safe margin on magnetic resonance imaging (MRI). The aim of this study was to assess which cut-off (1, 2 or 5 mm) was the best predictor of local recurrence based on preoperative MRI assessment of the circumferential resection margin (CRM). Data were collected prospectively on the distance between the tumour and mesorectal fascia for patients with documented radiological margin status in the MERCURY study. Positive margin and local recurrence rates were compared for MRI distances from the tumour to the mesorectal fascia of 1 mm or less, more than 1 mm up to 2 mm, more than 2 mm up to 5 mm, and more than 5 mm. The Cox proportional hazard regression method was used to determine the effect of level of margin involvement on time to local recurrence. Univariable analysis showed that, relative to a distance measured by MRI of more than 5 mm, the hazard ratio (HR) for local recurrence was 3·90 (95 per cent confidence interval 1·99 to 7·63; P < 0·001) for a margin of 1 mm or less, 0·81 (0·36 to 1·85; P = 0·620) for a margin of more than 1 mm up to 2 mm, and 0·33 (0·10 to 1·08; P = 0·067) for a margin greater than 2 mm up to 5 mm. Multivariable analysis of the effect of MRI distance to the mesorectal fascia and preoperative treatment on local recurrence showed that a margin of 1 mm or less remained significant regardless of preoperative treatment (HR 3·72, 1·43 to 9·71; P = 0·007). For preoperative staging of rectal cancer, the best cut-off distance for predicting CRM involvement using MRI is 1 mm. Using a cut-off greater than this does not appear to identify patients at higher risk of local recurrence.