Samples of raw and treated water after coagulation were collected from drinking treatment systems serving the cities of Akron, Barberton, Newton Falls and Ravenna (OH). Sampling was performed in a weekly basis (e.g., one to three samples each week) during periods comprising from two to three years, leading to the collection of between 600 and 1000 samples at each treatment facility. Water quality parameters (e.g., dissolved organic carbon (DOC), pH, ultraviolet absorbance at 254 nm (UV254)), bromide content, fluorescence excitation-emission matrices (EEM), and disinfection by-product and total organic halogen formation potential (DBPFP and TOX-FP) were determined for the samples before and after coagulation. Parallel factor analysis (PARAFAC) was applied in order to generate independent models on different subsets of each drinking water treatment plant (DWTP) data set: (i) raw water, (ii) treated water, (iii) composite data set (i.e., raw and treated water), and (iv) differential EEM ([delta]EEM)-based model. Three principal fluorophore groups were identified in the Akron, Barberton and Newton Falls raw and treated water data sets (two components with humic nature and a component with protein-like character), while four moieties (two humic-like and two protein-like components) were retained in the group of samples from Ravenna DWTP. Results of independent PARAFAC modeling were analyzed based on an uncorrected matrix correlation (UMC) approach in order to determine the impact of different coagulants on the structural character of the PARAFAC fluorophore groups. A quantitative analysis intended to study the distribution of the fluorophore moieties before and after treatment, predominant fluorescent structures in the treated water, and PARAFAC components being most affected by the specific coagulant in each DWTP was conducted. Results indicate that NOM in the water sources under monitoring has a highly similar spectral character. Principal conclusions after analysis in a multi-coagulant and multi-plant scenario included: (i) coagulation does not have a significant impact on the structure of the PARAFAC components, (ii) no new fluorescence entities are formed after coagulation, (iii) only physical removal of fluorophores is taking place in the coagulation process, and (iv) irrespective of the coagulant being applied (e.g., aluminum or iron-based salt), the same fluorescence entity (C2-high humic-like component) is the most affected by coagulation in terms of removal. PARAFAC analysis on [delta]EEM showed to be a valuable tool in order to determine recalcitrant fluorescence groups to coagulation treatment and to establish preferential removal of a specific moiety. Study of the coagulation process in the Akron DWTP, which corresponds to a parallel treatment train involving application of aluminum sulfate (alum) and aluminum chlorohydrate (ACH) on the same water source, confirmed that the fraction of NOM being impacted by these coagulants is identical and variations can only be noticed in the relative reduction attained for the estimated concentration of each fluorophore group in the NOM. Analysis of this particular DWTP demonstrated that a fluorescence-PARAFAC approach can improve the traditional DOC based-criterion used in DWTPs for selection and evaluation of a particular coagulant. Incorporation of PARAFAC components in a previously formulated semi-empirical coagulation model allowed establishing the role of each fluorophore group in the fraction of non-sorbable DOC (fraction of DOC that is not removed by coagulation) at each DWTP, offering improved understanding of the character of this organic material. Results showed that this fraction exhibited significant variation during the period of study at each treatment facility, while the fraction of sorbable DOC being effectively removed by coagulation had a significant non-linear association with the coagulant dose being applied; suggesting that marginal DOC removal will be attained after a specific concentration of coagulant has been applied. PARAFAC components showed to be suitable predictors of DBPFP and TOX-FP when multiple linear regression analyses were performed. Predictive capability differed for each set of raw and treated water samples and varied in an inter-DWTP basis. Higher association of PARAFAC components with trihalomethane formation potential (THMFP) was observed compared with the degree of fitting when the haloacetic acid formation potential (HAAFP) was analyzed. PARAFAC components with humic-like nature showed to be closely associated with THMFP and HAAFP, while structures with protein-like nature exhibited weak association with DBPFP and TOX-FP. PARAFAC analysis provided insight about the particularities of each water source and the efficiency of the specific treatment process applied in each facility. Results indicate that fluorescence analysis coupled with PARAFAC application may represent a practical tool to be used in the control and optimization of the water treatment operations increasing the efficiency of the processes (e.g., reducing chemical costs) and assuring the desired quality characteristics in the drinking water being supplied.