OBJECTIVE: The relationship between prostate-specific antigen (PSA) and prostate cancer (PCa) grade was traditionally thought to be linear but recent reports suggest this is not true in high-grade cancers. We aimed to compare the association between PSA and PCa-specific mortality (PCSM) in clinically localised low/intermediate and high-grade PCa. SUBJECTS/PATIENTS AND METHODS: Retrospective cohort study using the National Prostate Cancer Audit database in England of men treated with external beam radiotherapy (EBRT), EBRT and brachytherapy boost (EBRT + BT), radical prostatectomy or no radical local treatment between 2014 and 2018. Multivariable competing-risk regression was used to examine the association between PSA, Gleason, and PCSM. Multivariable restricted cubic spline regression was used to explore the non-linear associations of PSA and PCSM. RESULTS: 102,089 men were included, of whom 71,138 had low/intermediate-grade and 22,425 had high-grade PCa. In high-grade, 4-year PCSM was higher with PSA ≤5 than PSA 5.1-10 for men treated with EBRT (hazard ratio 1.96 (95% confidence interval 1.15-3.34) or no radical local treatment (hazard ratio 1.99 (95% confidence interval 1.33-2.98). Restricted cubic spline regression showed that PSA and PCSM have a non-linear association in high-grade but a linear association in low/intermediate-grade PCa. CONCLUSION: The low-PSA/high-grade combination in M0 PCa treated with EBRT has a higher PCSM than those with high-grade and intermediate PSA levels. In high-grade disease, the PSA association was non-linear; by contrast, low/intermediate-grade had a linear relationship. This confirms a more aggressive biology in low PSA secreting high-grade PCa and a worse outcome following treatment.