Petrochemical and agro-industries are facing similar issues for implementing robust calibration for on-line applications. Robust calibration methods like EPO and OSC-ED were developed by agro-industries to address the issue of calibration model robustness. Thus EPO and OSC-ED methods were tested in the framework of petrochemical applications. The potential and limitation of both methods were tested for improving the robustness of flash point prediction in presence of instrument disturbance (wavelength shift), sampling conditioning problem (temperature) and process problem (water pollution). OSC-ED corrected model is bringing some improvements for the predictability of model but the robustness was not significantly improved. On the other hand, EPO corrected model proved to be extremely robust while keeping a good predictability quality for the petrochemical robustness problem. Because EPO method is easily embedded in the calibration model, it can be easily implemented in the current NIR on-line systems. Potentially, with more robust model running on-line, EPO method can provide an elegant way to lower maintenance requirement regarding instrument, sampling system or process drift. In the same time, reliability of prediction from on-line NIR can be significantly increased.