OBJECTIVES: The Expected Value of Perfect Parameter Information (EVPPI) requires three level simulation in non-linear micro-simulation models, which is time consuming. Meta-modelling approaches reduce the time needed to evaluate the influence of uncertainty in complex health economic models. Here, two meta-modelling approaches are compared: spatial interpolation (SI) and ordinary least squares (OLS). METHODS: Both methods are applied to a drug-drug comparison in a micro-simulation schizophrenia model. The Net Benefits appear to depend rather non linearly on the underlying variables as assessed using the R2 and the RESET-test. Ordinary least squares is applied using 950 model runs. SI is applied using 100 and 950 model runs. Subsequently both meta-modelling methods are used to predict 50 out-of-sample runs and evaluated with the Root Mean Squared Prediction Error (RMSPE) and the time needed to estimate the EVPPI. RESULTS: One run of the micro-simulation model takes 11 minutes. The SI 950 is more accurate than the OLS and SI 100, which are comparable (average RMSPE for net benefits for two treatments 3006, 4842, and 4823 respectively). The SI 950 EVPPI calculation takes much longer time to complete the calculations; the SI 100 on the other hand is much quicker than the OLS (11005 minutes for the OLS and 2672 minutes for the SI 100). CONCLUSION: The SI is a frequentistic approach with great resemblance with the Bayesian Gaussian Process meta-modeling method. If a simulation model takes very long to come up with results or the model is non-linear, SI is the superior meta-modelling technique. If the model is linear, SI is fancy but not ideal.