The use of computer based prediction procedures to assist design is still virtually unknown. This paper argues that faced with the enormous complexity of typical design problems, the combustion engineer would be more than satisfied if he had access now to general prediction procedures of only qualitative precision. It contends that existing turbulence and combustion modelling can provide this. The need for computer economy is the controlling factor and the present lack of a procedure combining economy with geometric flexibility is only too apparent. The paper outlines two immediate priorities for combustion model improvement: the need for careful validation of manageable simulations of the effects of fluctuating density, and the necessity to allow for non-equilibrium of the overall reaction in diffusion flames. The latter requirement is supported by recent validation evidence.