In an attempt to reduce the effects of carbon dioxide emissions, there is a need for greater use of renewable energy sources, such as energy from biomass. In order to generate energy from biomass, the gasification process, by means of which it is possible to generate a noble fuel, can be highlighted. This work aimed to simulate the biomass gasification using artificial intelligence techniques, namely Artificial Neural Networks (ANN), using Matlab (trademark) software. Particularly, the objective was the development of ANN models with ten inputs (carbon, hydrogen, oxygen, nitrogen, volatile matter, moisture content, ash, equivalence ratio, temperature and steam/biomass ratio), applicable to a broad variety of biomass, with different types and concentrations of gasification agents in different types of gasifiers, capable of predicting the syngas composition (CO2, CO, CH4 and H2). Robust databases were built for training, testing and validation of the models, based on information collected in previous studies available in the literature and on the treatment of data obtained from the papers. Thirty-three neural network topologies were evaluated in order to choose the best one according to four criteria regarding training and test robustness. The network considered to have the best topology has 10 neurons in the input layer; 2 hidden layers, with logsig activation functions and 10 neurons in each hidden layer; the activation function is purelin in the output layer; 4 neurons in the output layer; and the training algorithm is trainbr. Such network has a good performance, with R2 values greater than 0.88 and 0.70 for training and test, respectively, for each of the four outputs. To evaluate the model, a validation was carried out, whose performance was not very appropriate, but it was possible to identify through a simple quantitative metric the more reliable regions where there is a greater density of training data.