Pyrosequencing targeting the V1-V3 hypervariable of the 16S rDNA was used to investigate the bacterial diversity in river and roof-harvested rainwater (RHRW) used for potable purposes by rural households in Luthengele village in the Eastern Cape Province of South Africa. The phylum Proteobacteria dominated the data set (80.5% of all reads), while 4.2% of the reads could not be classified to any of the known phyla at a probability of 0.8 or higher (unclassified bacteria). At class level, the classes; Betaproteobacteria (50.4% of all reads), Alphaproteobacteria (16.2%), Verrucomicrobiae (6.6%), Planctomycetacia (5.7%) and Sphingobacteria (3%) dominated the data set in all the samples. Although the class Verrucomicrobiae constituted 6.6% of all sequences, 88.6% of the sequences were from the river sample where the class represented 43.7% of the observed sequences in the sample. The bacteria community structure clearly showed significant similarities between RHRW and differences with the river water control sample, suggesting different levels of contamination and environmental factors affecting the various water sources. Moreover, signatures of potential pathogens including Legionella, Acinetobacter, Pseudomonas, Clostridia, Chromobacterium, Yersinia and Serratia were detected, and the proportions of Legionella were relatively higher suggesting a potential health risk to households using RHRW. This work provides guidance for prioritizing subsequent culturable and quantitative analysis, to ensure that potentially significant pathogens are not left out of risk estimations.