We aim to statistically analyse a large sample of results from diabetic retinal screening as graded by manual human graders and two different automated software programmes.