Mastering the production, transmission and distribution of electrical energy is a challenge that perpetually confronts electrical engineers. A precise short term electrical load forecasting results in economic cost savings and increased security in operating conditions, allowing electrical utilities to commit their own production resources in order to optimize energy prices as well as exchanges with neighboring utilities. The article is a review of the neural network based load forecasting techniques. We have based this review on relevant articles from the past four years. The purpose of the report is not to make an exhaustive bibliography, nor to make a comparison between different statistical forecasting techniques, but rather to stress and explain each technique with a few related articles. We have tried to explain the advantages and drawbacks of each technique. We also give some advice from our knowledge of the field, and advocate a common test protocol. Our description follows a main thread from the selection of the input variables, to the selection of the model and its estimation, and finally to the precision of the measure. Our description of the different systems covers the size of the input and the output layers of the models, and the type and number of models implied in the forecast