Organic matter (OM) is a ubiquitous constituent of natural waters quantifiable at very low levels using fluorescence spectroscopy. This technique has recognized potential in a range of applications where the ability to monitor water quality in real time is desirable, such as in water treatment systems. This study used PARAFAC to characterize a large (n=1479) and diverse excitation emission matrix (EEM) data set from six recycled water treatment plants in Australia, for which sources of variability included geography, season, treatment processes, pH and fluorometer settings. Five components were identified independently in four or more plants, none of which were generated during the treatment process nor were typically entirely removed. PARAFAC scores could be obtained from EEMs by simple regression. The results have important implications for online monitoring of OM fluorescence in treatment plants, affecting choices regarding experimental design, instrumentation and the optimal wavelengths for tracking fluorescent organic matter through the treatment process. While the multimodel comparisons provide a compelling demonstration of PARAFAC's ability to distill chemical information from EEMs, deficiencies identified through this process have broad implications for interpreting and reusing (D)OM-PARAFAC models.