Digital PCR is rapidly gaining interest in the field of molecular biology for absolute quantification of nucleic acids. However, the first generation of platforms still needs careful validation and requires a specific methodology for data analysis to distinguish negative from positive signals by defining a threshold value. The currently described methods to assess droplet digital PCR (ddPCR) are based on an underlying assumption that the fluorescent signal of droplets is normally distributed. We show that this normality assumption does not likely hold true for most ddPCR runs, resulting in an erroneous threshold. We suggest a methodology that does not make any assumptions about the distribution of the fluorescence readouts. A threshold is estimated by modelling the extreme values in the negative droplet population using extreme value theory. Furthermore, the method takes shifts in baseline fluorescence between samples into account. An R implementation of our method is available, allowing automated threshold determination for absolute ddPCR quantification using a single fluorescent reporter.