There seem to me to be four approaches to the problem of computing the evolution of star clusters. Firstly, one might assume that our knowledge of the evolution of stars can be condensed into a subroutine that can be added to an N-body code. This subroutine would mainly have to give the radius and the time-dependent mass of a star as a function of its initial mass and its age. Secondly, standing this on its head, one might assume that our knowledge of N-body evolution can be condensed into a subroutine that can be added to a stellar evolution code. This subroutine would determine, probably in a Monte-Carlo fashion, whether the star had picked up, or lost, a binary companion, or whether the orbit of its companion was significantly changed; the probabilities would be determined by simple analytic approximations to the time-dependent distribution functions of stars (and binaries) of different masses and ages, and by interaction cross-sections as functions of density and ‘temperature’. Thirdly, if the computing power is available, one might more simply unite an N-body code with a Stellar Evolution (SE) code, and follow both the dynamics and the internal evolution simultaneously. Fourthly, we might hope at some stage to put together simple analytic approximations both from N-body and from SE studies, to develop a unified simple model. I venture to say that it is only the last stage, if it is attainable, that would entitle us to say that we ‘understand’ the evolution of stellar clusters. ‘Understanding’, I think, means that we can extract some essential wisdom from large numerical simulations, and apply it on the back of the proverbial envelope.