Good control of air-fuel ratio under all operating conditions is essential for low exhaust emissions. In an effort to achieve this goal, an engine model based observer control structure has been applied to a single-cylinder CFR engine. The model includes fuel puddle dynamics, cycle delays inherent in the four-stroke engine process, and sensor dynamics for a universal exhaust gas oxygen (UEGO) sensor. This control structure has been shown to be capable of maintaining the air-fuel ratio within 0.5% rms of the commanded stoichiometric value during throttle transients. To achieve this level of performance, accurate values of model parameters such as time constants, delay times, and fuel puddle parameters are necessary. Since these parameters tend to vary with engine speed, throttle angle, time, and temperature, a method of periodically updating these parameter values is useful.This paper presents a nonlinear least squares parameter identification technique which provides accurate values of model parameters from data collected during normal engine operation. The inputs are the control signals for an electronic fuel injector and drive-by-wire throttle, while the output is the measurement from a UEGO sensor located in the exhaust manifold. Sensor characteristics and engine air flow parameters are identified during steady-state conditions by imposing small dithers about the nominal throttle angle, and fuel puddle parameters are identified during load transients. The identified values are then used in the model-based AFR control system previously described to control AFR over a range of throttle angles and engine speeds. Using this self-tuning regulator structure, air-fuel ratio control is demonstrated to follow the commanded stoichiometric value within 0.5& rms during various throttle transients with no off-line calibration effort.