The terminal velocity, v(infinity), of the stellar winds of 117 stars of spectral types O through F has been measured from the spectra in the new Atlas of Ultraviolet P Cygni Profiles by Snow et al. For the 68 stars with the most accurate determination of v(infinity) we studied the relation between v(infinity) and the photospheric escape velocity, v(esc). The ratio v(infinity)/v(esc), decreases from 2.58 +/- 0.20 in O-stars to 0.72 +/- 0.15 in A and F stars. There is a clear discontinuity in this ratio near T-eff similar or equal to 21,000 K at spectral type B1, where the ratio drops by almost a factor 2 from v(infinity)/v(esc) similar or equal to 2.6 at the high-temperature side v(infinity)/v(esc) similar or equal to 1.3 at the low-temperature side, and possibly a smaller discontinuity near T-eff similar or equal to 10,000 K. The discontinuity near 21,000 K is due to the change in the nature of the lines that produce the radiation pressure. At T-eff greater than or equal to 21,000 K the radiation pressure is due to high-ionization lines which are largely optically thick, whereas at T-eff less than or equal to 21,000 K the radiation pressure is due to very large numbers of low-ionization lines, which are more optically thin, Near T-eff similar or equal to 21,000 K the winds can exist in two states: winds with high mass-loss rates and low velocities, and winds with low mass-loss rates and high velocities. This is the bistability of stellar winds for stars with T-eff similar or equal to 21,000 K. We present evidence that the bistability may produce a jump in the mass-loss rates of the stars near T-eff similar or equal to 21,000 K with the higher mass-loss rates at the low temperature side of the jump. We briefly discuss the possible role of bistability in the formation of disks around B[e]-supergiants. The observed values of parameter of v(infinity) are used to derive empirical values of the force multiplier parameter alpha(emp) can be used to predict the terminal velocity of any early-type star with an accuracy of better than 5% with the method described in Appendix B, if the stellar parameters are known.
