Regression in the radial growth phase (RGP) of primary cutaneous melanomas is common and has been shown to be an adverse prognostic factor. However, the underlying mechanism is unclear. We performed dual immunohistochemical staining of podoplanin and S-100 on paraffin tissues from 321 patients with vertical growth phase primary melanomas, who had 10 years or more of follow-up. Lymphatic vessel density (LD) and lymphatic invasion (LI) were quantified and documented. The time to first metastasis and melanoma-specific death (MSD) from the date of definite treatment were analyzed using univariate and multivariate Cox models. Among the 116 vertical growth phase melanomas that had regression in the adjacent RGP, 75 (23%) were classified as complete and 41 (13%) were classified as partial. LD was significantly higher (P<0.001) in the 75 lesions with complete regression (mean±SD, 23.7±12.3/mm²) compared with the 41 lesions with partial regression (15.5±7.1/mm²) and was lower in 155 areas of the adjacent normal dermis (7.3±3.5/mm²) and 69 areas of the distant normal dermis (5.5±2.6/mm²). Patients whose lesions had areas of complete regression with LI and either high or low LD or had no LI with high LD, had shorter time to first metastasis (hazard ratio=2.5, 3.8, and 2.5, respectively) and increased risk of melanoma-specific death (hazard ratio=3.1, 1.3 and 3.0, respectively) than those with no LI, and low LD or those without areas of complete regression. These data indicate that complete RGP regression is associated with significantly increased LD. In addition, the adverse prognostic effect of RGP regression is at least partially mediated through lymphangiogenesis and LI in this area.