Abstract It is commonly the case in biochemical modelling that we have knowledge of the qualitative ‘structure’ of a model and some measurements of the time series of the variables of interest (concentrations and fluxes), but little or no knowledge of the model's parameters. This is, then, a system identification problem, that is commonly addressed by running a model with estimated parameters and assessing how far the model's behaviour is from the ‘target’ behaviour of the variables, and adjusting parameters iteratively until a good fit is achieved. The issue is that most of these problems are grossly underdetermined, such that many combinations of parameters can be used to fit a given set of variables. We introduce the constraint that the estimated parameters should be within given bounds and as close as possible to stated nominal values. This deterministic ‘proximate parameter tuning’ algorithm turns out to be exceptionally effective, and we illustrate its utility for models of p38 signalling, of yeast glycolysis and for a benchmark dataset describing the thermal isomerisation of α-pinene.