In order to realize the online evaluation of diesel engine lubricating oil condition, the variation laws of dielectric constant and viscosity with the main deterioration indicators such as diesel content, soot content and oxidation degree were acquired. The diesel content had little effect on the dielectric constant and showed an obvious negative correlation with viscosity. The soot content had a positive correlation with the dielectric constant and a quadratic relationship with the viscosity. The influence of oxidation degree on both could be ignored. Then an evaluation model of the main deterioration indicators for lubricating oil was established based on regression analysis and BP neural network. The experimental results show that BP neural network model has a higher prediction accuracy and the maximum prediction errors for main deterioration indicators are 8.3%, 9.7% and 6% of their respective content threshold. The online real-time monitoring of diesel engine lubricating oil by the built model can avoid damage to engine parts caused by abnormal state of lubricating oil and it is of great significance to improve the reliability of diesel engine.