A performance evaluation of the detection and classification algorithm for earthquake early warning was conducted to test its reliability and robustness. The Seismic Alert System of Mexico (SASMEX) has used this algorithm since 1991. The algorithm estimates the rate of seismic energy released during two times the () period. Based on the energy released, it estimates an empirical magnitude range related to . Depending on the estimated , either preventive or public alerts are issued. In this article, post facto tests are presented for 61 earthquakes for which SASMEX issued an alert. The algorithm was also tested on 31 earthquakes () that occurred in the Mexican subduction zone from 1985 to 2014. These earthquakes occurred outside the coverage of the SASMEX instruments at the time. This dataset includes the 19 September 1985 8.1 Michoacán earthquake and the 9 October 1995 8.0 Colima event. The algorithm was tested also on two great earthquakes: the 22 February 2010 8.8 Maule, Chile, earthquake and the 11 March 2011 9.0 Tohoku, Japan, event. The results of the evaluation of 144 acceleration records of the 61 earthquakes detected from the SASMEX network indicate that 92% of the accelerograms of earthquakes with have errors in the prediction of magnitude of less than , and 83% for . Also, the tests conducted on the 59 acceleration records of 31 earthquakes with indicate that in all cases, with the exception of one strong‐motion record, the events are classified as . Thus, the algorithm shows a high level of reliability and robustness. Although the algorithm underestimates the magnitudes of large earthquakes, these events are identified and classified as . Thus, an alert would be issued for these great earthquakes.