We present a study for optimization of a combustion process based on optical measurements. First, we adapt a data-driven Hammerstein model on optical and key variables in a laboratory scale boiler. We then discuss the concepts of combustion diagnosis, optimization based on flame spectrum analysis and the inclusion of optical variables in the control system. We compare boiler manual operation, conventional optimization method based on the boiler energetic performance index and advanced boiler optimization methods based on optical variables, including empirical information related to pollutants emissions. The most relevant result is a reduction of 13% in the average carbon monoxide emission, and a similar reduction in carbon monoxide emission as function of time during the optimization algorithm start-up, with a marginal reduction of less than 1% on the boiler energetic performance index.