The UV-persulfate oxidation method is widely used for determining the total organic carbon concentration of aqueous samples (denoted for convenience as UVP-TOC). However, for some surface water samples, the measurement of TOC by this method can be unreliable, deviating significantly from the true carbon content. In this study, the performance of the UVP-TOC method has been investigated by comparing the results from the analysis of a variety of aqueous samples that included two kinds of surface water samples and related surface water model substances: bovine serum albumin (BSA), sodium alginate (SA), humic acid (HA), tannic acid (TA), benzoic acid (BA) and citric acid (CA), with those from a high-temperature combustion method (elemental analysis); the latter providing the true carbon content value. By comparing the above data, it was found that the UVP-TOC method significantly underestimated the TOC value of the surface water samples, and it was also found that the model components BSA (protein) and HA (humic substances, HS) had a substantial influence on the TOC underestimation, while the SA (polysaccharide), TA (complex organic molecule) and CA/BA (small molecules) had little effect. The results showed that the agglomeration within and between BSA and HA molecules was an important reason for the inaccurate UVP-TOC values of BSA and HA. A further limitation was that for BSA, surfactants (e.g. sodium dodecylbenzene sulfonate, SDBS) and other surfactant-like substances, foam was formed during the CO2 removal purging process by N2 that seriously interfered with the determination of TOC. The study provides new information and insight into the causes of inaccuracies in the UVP-TOC analysis of surface waters and possible approaches to improve the accuracy.