A variant form of Clostridium perfringens epsilon toxin (Y30A-Y196A) with mutations, which shows reduced binding to Madin–Darby canine kidney (MDCK) cells and reduced toxicity in mice, has been proposed as the next-generation enterotoxaemia vaccine. Here we show that, unexpectedly, the Y30A-Y196A variant does not show a reduction in toxicity towards Chinese hamster ovary (CHO) cells engineered to express the putative receptor for the toxin (myelin and lymphocyte protein; MAL). The further addition of mutations to residues in a second putative receptor binding site of the Y30A-Y196A variant further reduces toxicity, and we selected Y30A-Y196A-A168F for further study. Compared to Y30A-Y196A, Y30A-Y196A-A168F showed more than a 3-fold reduction in toxicity towards MDCK cells, more than a 4-fold reduction in toxicity towards mice and at least 200-fold reduction in toxicity towards CHO cells expressing sheep MAL. The immunisation of rabbits or sheep with Y30A-Y196A-A168F induced high levels of neutralising antibodies against epsilon toxin, which persisted for at least 1 year. Y30A-Y196A-A168F is a candidate for development as a next-generation enterotoxaemia vaccine.