A fuel mass estimation method for short pilot diesel injections is proposed and analyzed in this article. Previous studies showed that the pilot misfire ratio was more strongly correlated with the fuel mass than the on-time. This characteristic is exploited for the fuel mass estimation in a region where it is otherwise challenging to get good estimation accuracy due to the low signal-to-noise ratio, such as by rail pressure measurements or in-cylinder pressure for heat release estimation. The suggested method uses a Bayesian approach where the calibrated injectors, the pilot misfire ratio and the misfire detection are stochastically modelled. The effect of the different model parameters and dispersion on the estimator properties are analyzed. Experimental results in a Scania D13 Diesel engine confirm the improvement in the pilot mass estimation, for the regions within the transition from full misfire to full combustion. In this region, a 60% reduction in the estimation error was obtained, from 0.66mg to 0.27mg standard deviation.