We trace the failure of Einstein causality in interacting high-spin field theories to the presence of secondary constraints in the Cauchy initial-value problem. We bring in and extend a theorem of Gel'Fand and Yaglom, and use it to show that non-negativity of the field anticommutators for half-integral spin, and non-negativity of the field energy for integral spin, necessarily imply the existence of such constraints, except for the privileged spin values 0, 4 and 1. We prove this under very lax assumptions as to the nature of the mass spectrum.We carry out a systematic search for alternative spin-; constraint schemes, taking a Lagrangian which contains thirteen adjustable parameters, many of which couple in the hitherto 'obscure' unmixed-spinor representations (a, 0) and (O,$ of the Lorentz group.The search reveals two unfamiliar spin-; equations, one of which is entirely new. We analyse the new equations and find that each contains a Rarita-Schwinger constraint subchain. This means that the new equations will also show causality failures.We find that in five or more space-time dimensions additional field equations appear, which are very much simpler and give a much more fundamental role to the 'obscure' representations. We infer that our inability to find a viable alternative to the Rarita-Schwinger scheme reflects an essential mathematical feature of four-dimensional spacetime, and that it is not due just to a lack of imagination. We examine the dimension dependence of the compound-spin equation of Harish-Chandra, and find further support for this assessment. We conclude that the only acceptable dynamical equations in fourdimensional Minkowski space-time are those customarily associated with spins 0, + and 1.