In recent years, there has been an increasing interest for the use of Principal Component Analysis for the development of reduced-order combustion models. The PCA modelling framework has been demonstrated a priori and a posteriori for a wide range of configurations, including simple batch and perfectly stirred reactors, one-dimensional laminar flames and complex cases such as flame-vortex interaction as well as plasma flows. Results indicated that PCA-based models are able to provide very accurate results when compared to full size simulations. Current investigations have shown that PCA models are relatively invariant to parameters such as the Reynolds number of the flow. This implies that PCA models can be trained on relatively simple systems and used to simulated systems showing more complex turbulence/chemistry interactions. However, it appears still necessary to determine and quantify the validity range of PCA models, to determine: i) the required degree of complexity of the chemical reactor used to generate the model; and ii) the conditions at which the reduced models will not be valid anymore.