In an evaluation of diatoms as indicators of human disturbance in coastal ecosystems of the Laurentian Great Lakes, we characterized assemblage specificity to lake and habitat type to identify non-anthropogenic factors influencing indicator models. Surface sediment assemblages and environmental variables were collected along the U.S. coastline at 191 sample sites, which were classified by lake and geomorphic type: high-energy (HE), embayment (EB), coastal wetland (CW), riverine wetland (RW), protected wetland (PW), and open water (OP). Diatom inferred (DI) total phosphorus (TP) transfer functions (models) were developed for each lake and geomorphic type. Robust models included: the overall model (RMSEP; r²_jack = 0.65; RMSEP = 0.005), Lake Superior (r²_jack = 0.73; RMSEP = 0.003), Lake Ontario (r²_jack = 0.73; RMSEP = 0.007), PW (r²_jack = 0.64; RMSEP = 0.003), and EB (r²_jack = 0.64;RMSEP = 0.007). Weaker models, indicating poorer diatom-TP relationships, included: RW (r²_jack =0.03; RMSEP = 0.005), OP (r²_jack = 0.15; RMSEP = 0.059), and Lake Michigan (r²_jack = 0.38; RMSEP =0.006). DI TP data were regressed against landscape characteristics to quantify the relationships to adjacent watershed stressors. RW data were further scrutinized as a case study to investigate the suitability of diatom-based approaches in systems with poor diatom-TP relationships. Despite poor performance of the RW model, DI phosphorus data for riverine wetlands, derived from the overall model, were strongly related to watershed characteristics (r² = 0.61), indicating the overall model's ability to integrate stressors from the surrounding watershed in areas where measured phosphorus did not adequately characterize prevailing conditions. This study confirms that physical properties (e.g., lake or habitat type) can influence indicator models; however, weaknesses may be overcome by robust calibration techniques.