Based on the radial based function neural network (RBFNN) and the orth o gonal least square (OLS) algorithm, an online diagnostic strategy for sensor f a ult detection of the cold start diesel engine is proposed. The strategy for sens or fault detection of the cold start diesel engine is realized by using the sens or sampling data as input of the RBFNN and the sensor faults as the output of th e RBFNN to train the network. The online diagnostic tests for the sensor hardw a re malfunctions such as short circuit, open circuit and the fixed value of the e lectric current sensor, the voltage sensor and the rotate speed sensor, and also the software malfunctions such as the errors from linearity, the sensitivity an d the repeatability are made on the cold start diesel engine by using the RBFNN and the OLS algorithms. The results show that the diagnostic accuracy can reach 956％, the maximal linearity error is 05％, the maximal sensitivity error is 08％, and the maximal repeatability error is 01％.