This paper presents the measured frequency diversity improvement factor for 6 GHz 16-QAM 90 Mbit/s digital radio on the 26.4 mi Atlanta-Palmetto path in Georgia. Two channels with a center frequency separation of 59.3 MHz were used in a one-by-one frequency diversity experiment. The 1980 data and the 1982 data indicate a frequency diversity improvement factor of 100 and 45, respectively, at the outage threshold of 10-3BER. This is in contrast to the improvement factor of 9 predicted for analog FM radio at the same fade margin. The measured one-by-one frequency diversity improvement factor is comparable to the measured space diversity improvement factor with 30 ft antenna spacing on the same path. We conclude that 1) frequency diversity can provide a large improvement factor for digital radio, 2) as an alternative to space diversity, frequency diversity can provide substantial cost savings for digital radio routes, and 3) the frequency diversity calculation based on analog FM radio experience is too conservative (i.e., pessimistic) for digital radio application. These experimental findings are in agreement with recent advances in digital radio diversity modeling. Digital radio performance depends heavily on the multipath dispersion in the channel. The measured data indicate that the power fade depths in the two channels are highly correlated, whereas the multipath dispersion in the two channels is decorrelated. This correlation difference provides insight into the measured large frequency diversity improvement factor for digital radio.