An adaptive neuro-fuzzy inference based delay estimation system is proposed. The system is compared with other delay estimation models, and tested through simulation and observation values. Rules, fuzzification and inference are modeled by neuro-fuzzy. Hybrid algorithm has been used for training and tests. The rule base of the delay estimation system is constructed either following a mathematical model or from real-time traffic operational data. This study has shown that adaptive neuro-fuzzy technique, a method to predict vehicle delays at signalized junctions, can be successfully applied to modeling of traffic systems. Introduction In order to minimize vehicle delays on roads and evaluate alternative junction construction projects, the amount of delays should be estimated with high accuracy. There are numerous works 1-3 that employ fuzzy logic and artificial neural networks on signal control and vehicle delay estimation at signalized junctions. Neuro-fuzzy systems profit from both the linguistic, human-like reasoning of fuzzy systems and the powerful computing ability of neural networks 4-13 . This study develops a neuro-fuzzy logic based system using a mathematical model in a simulation study for estimating vehicle delay at a junction.