Background: Only a subset of advanced melanoma patients respond to anti-PD-1 (aPD1) monotherapy. Upfront identification of (non-)responsiveness would help guide first-line treatment decisions, prevent overtreatment and unnecessary risk for toxicities. T cell density and expression of T cell related genes have been associated with response to aPD1, but are imperfect predictors. We investigated whether spatial proximity of CD8 T cells to tumor cells improves upon the predictive value of T cell density alone. Methods: Pretreatment tumor specimens from melanoma patients treated with aPD1 in the Netherlands Cancer Institute were stained for DAPI, SOX10/Melan-A, CD4, CD8, FOXP3 and PD-1 by multiplex immunofluorescence. Sections were imaged on Vectra and analyzed using HALO to optimize marker thresholds and demarcate tumor and stroma. T cell proximity to tumor cells was evaluated as difference in area under the curve between i) a spatial G-function quantifying T cell density around tumor cells in tumor areas and ii) analogous null distributions obtained by random permutation of cell labels. This assessment of co-clustering is independent of cell density and heterogeneity therein and does not reflect repulsion of T cells to stromal/marginal areas. Clinical characteristics, RECIST response and survival were collected from patient records. Associations between T cell density, T cell proximity to Sox10/Melan-A⁺ tumor cells, other clinical biomarkers (LDH, M stage and WHO performance status) and response were examined in a Bayesian hierarchical logistic regression. Results: Tumor specimens of 98 patients were included, of whom 45 were treated with aPD1 as first-line therapy and 33 had an objective response. CD8 T cell proximity to tumor cells was associated with response in an independent, comparatively strong, and tissue dependent manner (cutaneous tissue: 2.78 [2.45, 3.17], visceral: 2.30 [1.95, 2.72], lymphoid: 2.12 [1.88, 2.40], format: maximal posteriori odds ratio [89% equal-tailed credibility interval]), in a multivariate model correcting for CD8 T cell density (1.74 [1.62, 1.88]), LDH (1.93 [1.72, 2.16]), M stage (0.92 [0.87, 0.98]) and WHO performance status (0.79 [0.72, 0.88]). Our model achieved an area under the ROC curve of 77.7%, whereas an analogous model omitting the proximity variable achieved 73.1%. Conclusions: Our analyses show that spatial proximity of CD8 T cells to tumor cells functions as an independent biomarker for response to aPD1 and suggests that preexisting CD8 T cell tumor reactivity is reflected by this spatial proximity.