Merkel cell carcinoma (MCC) is an aggressive cutaneous malignancy that is predisposed to local recurrence and regional lymph node (LN) metastases. Current nodal staging for MCC is based on the location of metastatic LNs (regional versus in-transit) and whether a metastasis is detected macroscopically or only microscopically. However, this may underestimate the cumulative effect of increasing metastatic nodal burden, which is a primary driver of survival in multiple malignancies. To this end, we examined the independent, quantitative impact of metastatic LN burden on survival in a large MCC cohort. Patients with non-metastatic MCC undergoing upfront surgical resection with pathologic LN staging between 2004 and 2014 were identified in the National Cancer Database. Univariate and multivariate models were constructed to evaluate the association between survival and number of metastatic LNs, adjusting for patient, tumor, and demographic factors. Restricted cubic spline functions were used to model the non-linear relationship between number of positive LNs and survival. Recursive partitioning analysis (RPA) was used to derive a novel MCC nodal classification system based on nodal covariates, including number of LNs and regional versus in-transit metastasis. Overall, 4218 patients met inclusion criteria, including 1817 (43%) with at least 1 positive LN and 2711 (57%) with no metastatic LNs. In multivariable models, mortality risk increased continuously without plateau with increasing number of metastatic LNs. The risk of death increased most rapidly for the first 2 positive LNs, (hazard ratio [HR] per LN, 1.53; 95% CI, 1.40-1.67; P<0.001), and continued to increase with each positive LN beyond this (HR per LN, 1.04; 95% CI, 1.02-1.06; P<0.001). RPA identified 4 distinct clusters of patients that were used for a new nodal classification system: 0 LNs (N0), 1-2 LNs (N1), 3-8 LNs (N2), and ≥9 LNs (N3). 5-year overall survival for these patients was 67.2%, 46.8%, 30.2%, and 11.2%, respectively. In multivariable Cox regression, the HR for death in comparison to N0 patients was 1.91, 3.41, and 5.94 (all P<0.001) in RPA-derived nodal classifications N1, N2, and N3, respectively. In comparison, the respective HR for American Joint Committee on Cancer (AJCC) nodal classifications N1a, N1b, and N2 were 1.43, 2.69, and 2.86 (all P<0.001) versus N0 patients. Incorporating the RPA-derived system based on LN number into a model with other patient, tumor, and demographic factors exhibited greater concordance with survival than the AJCC system. The number of metastatic LNs is a primary factor influencing survival in MCC. An RPA-derived nodal classification system based solely on this variable improves the accuracy of patient stratification and partitions patients across a wider spectrum of mortality risk in comparison to the current AJCC system. Thus, metastatic LN number may be used to refine MCC staging and help direct adjuvant treatment recommendations.