After the PR disaster of 1997 XF11 (March 1998), we started a crash research program on impact predictions. The difficulty was the chaotic motion of Earth-crossing asteroids (orbit uncertainty increases exponentially with time); it can be solved by replacing a real asteroid with a swarm of Virtual Asteroids. In 1999 we introduced Geometric Sampling to replace Monte-Carlo methods (see Milani, Chesley & Valsecchi, A&A 346, 1999). In November 1999 the first Impact Monitoring system CLOMON was operational. From 2002 the second generation systems CLOMON2 at Pisa and SENTRY at JPL are operational: critical cases are scanned for possible impacts in the next 90–100 years.