[1] OH and HO2 mixing ratios and total OH reactivity were measured together with photolysis frequencies, NOx, O3, many VOCs, and other trace gases during the midsummer 1999 SOS campaign in Nashville, Tennessee. These measurements provided an excellent opportunity to study OH and HO2 (collectively called HOx), and their sources and sinks in a polluted metropolitan environment. HOx generally showed the expected diurnal evolution, with maxima around noon of up to about 0.8 pptv of OH and 80 pptv of HO2 during sunny days. Overall, daytime observed OH and HO2 were a factor of 1.33 and 1.56 times modeled values, within the combined 2σ instrument and model uncertainties. The chain length of HOx, which is determined from the ratio of the measured total OH reactivity that cycles OH to the total HOx loss, was on average 3–8 during daytime and up to 3 during nighttime, in general agreement with expectations. However, differences occurred between observed HOx behavior and expectations from theory and models. First, HO2 was greater than expected during daytime when NO mixing ratios were high; ozone production did not decrease as expected when NO was greater than 2 ppbv. Ozone production determined by the imbalance of the NOx photostationary state, which was almost twice that from HO2, also shows this dependence on NO. Second, the calculated OH production rate, which should equal the measured OH loss rate because OH is in steady state, is instead less than the measured OH loss rate by (1–2) × 107 molecules cm-3 s-1, with low statistical significance during the day and high statistical significance at night. Third, surprisingly high OH and HO2 mixing ratios were often observed during nighttime. The nighttime OH mixing ratio and the HO2/OH ratio cannot be explained by known reaction mechanisms, even those involving O3 and alkenes. Because instrument tests have failed to reveal any instrument artifacts, more exotic chemicals or chemistry, such as OH adducts or other radicals that fall apart into OH inside the instrument, are suspected.