The Norwegian national landslide early warning system (LEWS), operational since 2013, is managed by the Norwegian Water Resources and Energy Directorate and was designed for monitoring and forecasting the hydrometeorological conditions potentially triggering slope failures. Decision-making in the LEWS is based upon rainfall thresholds, hydrometeorological and real-time landslide observations as well as on landslide inventory and susceptibility maps. Daily alerts are issued throughout the country considering variable size warning zones. Warnings are issued once per day for the following 3 days and can be updated according to weather forecasts and information gathered by the monitoring network. The performance of the LEWS operational in Norway has been evaluated applying the EDuMaP method, which is based on the computation of a duration matrix relating number of landslides and warning levels issued in a warning zone. In the past, this method has been exclusively employed to analyse the performance of regional early warning models considering fixed warning zones. Herein, an original approach is proposed for the computation of the elements of the duration matrix in the case of early warning models issuing alerts on variable size areas. The approach has been used to evaluate the warnings issued in Western Norway, in the period 2013–2014, considering two datasets of landslides. The results indicate that the landslide datasets do not significantly influence the performance evaluation, although a slightly better performance is registered for the smallest dataset. Different performance results are observed as a function of the values adopted for one of the most important input parameters of EDuMaP, the landslide density criterion (i.e. setting the thresholds to differentiate among classes of landslide events). To investigate this issue, a parametric analysis has been conducted; the results of the analysis show significant differences among computed performances when absolute or relative landslide density criteria are considered.