Radiation myelopathy (RM) is an uncommon but serious late effect of thoracic radiotherapy (RT), which oncologists try to avoid by careful planning and dose selection. Five patients with RM are described from among 1048 with inoperable non-small cell lung cancer treated with palliative RT in three randomized trials conducted by the Medical Research Council Lung Cancer Working Party. Seven RT regimens were used in these trials: 10 Gy in a single fraction on one day (10/1/1) (114 patients), 17/2/8 (524 patients), 27/6/11 (47 patients), 30/6/11 (36 patients), 30/10/12 (88 patients), 36/12/16 (86 patients) and 39/13/17 (153 patients). Of the five instances of RM, three occurred in the 524 patients treated with 17 Gy in two fractions, and two in the 153 treated with 39 Gy in 13 fractions. The estimated cumulative risks of RM by 2 years were 2.2% for the 17 Gy group, 2.5% for the 39 Gy group, and 0% for the remainder, but the annual risks had wide 95% confidence intervals, indicating that the distribution of episodes among the seven regimens could have been random. Nevertheless, calculation of cord doses in terms of the total doses that would have an equivalent biological effect if given in 2 Gy fractions (LQED2 values) from our data for different values of the ratio of the linear quadratic parameters of the cell survival curve (alpha/beta), suggest that the best estimate of alpha/beta is less than 3 Gy, and possibly close to 2 Gy. This emphasizes the sensitivity of human spinal cord to changes in fraction size. We recommend that, when the computed LQED2 for a schedule of treatment that includes the thoracic spinal cord (assuming alpha/beta = 2 for cord) exceeds 48 Gy, oncologists should consider reducing the dose to the cord.