The concern over extensive pollution, including anthropogenic carbon dioxide emission caused by the use of fossil fuels, results in the transition of the fuel mix of the world toward renewable energy sources. One of the most promising biofuels is biodiesel, which is renewable, nontoxic, biodegradable, safe to store, handle, and transport, and produces lower pollutant emissions (except oxides of nitrogen) compared to fossil diesel. However, one of the potential problems associated with biodiesel is the variability in its fatty acid methyl ester composition owing to larger variations in the feedstock used for its production. The biodiesel composition variations leads to variations in fuel properties, and thereby engine characteristics, demanding engine recalibration every time a new biodiesel fuel is introduced. In the present study, biodiesel-composition-based models are developed using artificial neural networks (ANN) to predict combustion, performance, and emission characteristics of a light duty naturally aspirated and a heavy duty turbocharged engine fuelled with different types of biodiesel. The models provide predictive functions for estimating the engine performance, combustion, and emission parameters across a range of biodiesel composition, thus reducing extensive engine experiments. The predictions from the developed ANN models compare well with measurements with a higher regression coefficient of above 0.9 and less than 10% absolute error. Further, attempts are made to combine the developed ANN models with a genetic algorithm to arrive at an optimal biodiesel composition which could result in better fuel economy and lower oxides of nitrogen emission. The obtained results show that the total saturated methyl ester falls in the range of 36-43% by weight and that the total unsaturated methyl ester falls in the range of 55-63% by weight for the optimum biodiesel composition.