The variability of small-scale atmospheric turbulence on time scales from a few minutes to the annual scale and at altitudes from about 4 to 20 km is studied using clear air Doppler radar data from Poker Flat, Alaska, and Platteville, Colorado. The variable used for this study is the refractivity turbulence structure constant Cn2. The frequency distribution of Cn2 is found to be lognormal during all seasons at all altitudes. Using data at 4-min intervals, it is found that the autocorrelation function of Cn2 can be modeled as the sum of a first-order autoregressive process and a random process. The associated integral time scale decreases with altitude from 25 to 45 min in the troposphere to about 18 min in the stratosphere. The power spectrum of log Cn2 follows a power law relation with frequency; at periods greater than about 2 hours the spectral slope is near -5/3 and at periods less than 2 hours the slope is near -1. Monthly mean values of log Cn2 are largest in the winter and show a secondary maximum in summer. The winter peak is apparently related to increased jet stream and baroclinic strom activity, and the summer peak is believed to be due to convective activity. The correlation of 3-hour values of Cn2 with wind speed over monthlong periods ranges as high as 0.8 and has a median value near 0.3. During certain periods, Cn2 also depends on other variables such as boundary layer inversions and gravity wave activity. Monthly mean values of Cn2 correlate more closely with an indicator of the intensity of gravity wave activity than with wind speed or the coarse-scale wind shear. The meteorological significance of these results is discussed briefly.