Background: Patients with primary cutaneous melanoma are selected for sentinel lymph node biopsy (SLNB) based upon their risk of a positive SLN being identified. To improve upon stage-based categorization, the Memorial Sloan Kettering Cancer Center (MSKCC) and Melanoma Institute Australia (MIA) developed predictive models. Which model is more clinically useful is currently unclear. Methods: Consecutive patients undergoing SLNB from 2007-2021 were identified from the Swedish Melanoma Registry (SweMR), which covers 99% of patients with invasive cutaneous melanoma in Sweden. The predicted probability of SLN positivity was calculated using the MSKCC and limited MIA models (using mitotic rate as absent/present instead of count and excluding the optional variable lymphovascular invasion) for each patient. The operating characteristics of the models were assessed and compared. The “clinical utility” of each model was assessed using decision curve analysis (DCA) and compared with a strategy of performing SLNB on all patients. Results: In total 10,089 patients were included; the median Breslow thickness was 1.8 mm, 33.7% were ulcerated, and 1,802 patients had a positive SLN (17.9%). Both the MSKCC and limited MIA models were well calibrated across the full range of predicted probabilities, and had similar external area under curve (AUC), MSKCC 70.8% (95% CI 69.5-72.1%) and MIA 70.3% (95% CI 68.9-71.7%). When new models were constructed to compare the incremental value of the additional parameters of either nomogram over Breslow thickness alone, the AUCs were 70.1%, 71.6% and 72.0% for Breslow alone, MIA parameters and MSKCC parameters, respectively. At a risk threshold of 5%, DCA indicated no added net benefit for either model compared to performing SLNB for all patients. At risk thresholds of 10% or higher, both models added net benefit compared to SLNB for all. The greatest benefit was observed from use of the models in patients with T2 melanomas when a threshold of 10% was used to select patients for SLNB. In that setting, use of the nomograms led to a net reduction in seven avoidable SLNBs per 100 patients for the limited MIA nomogram and eight for the MSKCC nomogram, when compared to a strategy of SLNB for all. Conclusions: This study confirms the statistical performance of both the MSKCC and limited MIA models in a large, nationally-representative dataset. However, clinical utility as assessed by DCA demonstrated that using the models only improved selection for SLNB when a risk threshold of at least 10% was employed, and then mainly for T2 melanomas. Either further development of the nomograms, or other strategies, are needed to improve selection at lower risk thresholds.