The receiver operating characteristic curve is a simple graphical tool used to assess the accuracy of diagnostics tests. Pulit (2016) proposed an innovative approach for estimating the receiver operating characteristic curves based on kernel smoothing. Although his proposed estimator is highly appealing in several aspects, it suffers from the well-known boundary bias effect. In this paper, we highlight this drawback and propose a new modified estimator that uses an appropriate boundary kernel. The asymptotic convergence of the proposed estimator at boundary points is demonstrated. Using both simulated and real data sets, we illustrate the performance of the proposed estimator. The results show that the proposed estimator outperforms not only the Pulit’s estimator but also other commonly used estimators.