Impulsive Pi2 pulsations have long been recognized as one of the key signatures of magnetic activity during substorm periods due to their wide observable range both in latitudes and longitudes. It is well documented that there is usually more than one Pi2 wave burst associated with a substorm and only one of them corresponds to the onset of the substorm. This observational fact poses obstacles to determining substorm onsets with Pi2 signals. Although the Pi2 have become one of the most popular indicators for substorm onsets, the reliability of using the Pi2 in this fashion has not been seriously investigated. In this paper we address this question with a statistical approach by using ~650 auroral substorm onsets identified with Polar ultraviolet images for a time interval from April 1996 to May 1997. A comparison of the low-latitude Pi2 pulsation onsets from Kakioka (L=1.07) with the auroral breakups indicates that identifying substorm onset with the Pi2 alone is often ambiguous. Of a total of 119 isolated (defined as ~10 min of quiet time preceding the onset) Pi2 bursts seen within ~10 min from a magnetic positive bay, there were 65 events (~55%) taking place within 3 min from breakups and 34 events (29%) indicating no sign of an auroral breakup within 10 min of the Pi2 burst. This result suggests that Pi2 may not be as a good indicator of the substorm onset as it was thought to be. Interestingly, it is always possible to associate Pi2 pulsations with some forms of auroral intensification. When compared to auroral breakups, Pi2 onsets are subject to a small delay of 1--3 min, with a peak around 1 min. Delays of Pi2 onsets are revealed to be a function of location relative to auroral breakup. This dependence is found to be consistent with the time of flight for a fast-mode wave, in a plasmapause cavity mode model, propagating in the magnetosphere. ¿ 2000 American Geophysical Union