Flight fuel is one of the most important and cost-intensive factors for airlines in terms of both safety and profitability. Accurate estimation of the required amount of trip fuel to be loaded in an aircraft is therefore a major challenge that requires sophisticated models and a wide range of operational parameters. However, current approaches often deviate from the actual performance of a flight due to various factors such as planned weather conditions or cruise level. This article introduces a self-organizing constructive neural network (CNN) that determines connection weights analytically and thus improves trip fuel estimation. Moreover, it is shown that the proposed model outperforms an existing BPNN in terms of accuracy and complexity.