This work aims at presenting a method for heartbeat classification based on multi-layer perceptron (MLP) and random forest (RF) techniques applied on the first difference of ECG signals. From the MIT-BIH Arrhythmia Database, each annotated P-QRS segment was extract, low-pass filtered, and first-order difference be used as input of the neural networks. The MLP and Random Forest was used to obtain a model for classifying the heartbeats. The results where compare with other algorithms existing in the literature, and the model developed attached good results and notice improvements when used the first difference.