The Viking infrared thermal mapper (IRTM) has measured reflected and emitted energy over Mars south polar cap throughout the martian spring and summer. During these 1976-1977 observations the polar cap displayed complex spatial, spectral, and temporal variations. The cap recession was accompanied by a bright rim and dark internal collar. High albedos did not directly correspond with low tempertures as would be expected for a simple polar frost model. Both the polar cap and the surrounding terrain brightened erratically during the global dust storms. The polar cap remained approximately symmetrical around the geometric pole until about Ls 237, when the eastern hemisphere )longitudes 180¿-360PmD) darkened and the western hemisphere began to develop spectral contrasts at thermal wavelengths. As the cap shrank to its residual position centered at -87.5¿, 30¿W, it maintained large differences in brightness temperature between the four IRTM surface-sensing bands at 7, 9,11 and 20 μm. Whereas T9 reached 214 K, T20 never exceeded 175 K. No single mechanism explains the observed spectral behavior. The late summer infrared spectral pattern can be matched by a surface consisting of CO2 frost with 20-μm emissivity of 0.8 and ~6% dark, warm soil under a dusty atmosphere of moderate infrared opacity and spectral properties similar to those measured for the martian global dust storms. Throughout the summer, the polar frost remained at the temperature of solid CO2. This low temperature, the spectral pattern, the absence of appreciable water vapor in the south polar atmosphere, and the absence of surface warming expected if H2O ice were to become exposed, all imply that the residual south polar cap seen by Viking was covered by solid CO2. Thus Mars appears to have a residual polar cap of CO2 in the south and one of H'd2O in the north. This difference is due in part to the occurrence of global dust storms during the southern summer. An accumulation of solid CO2 over the location of the residual cap beyond that predicted by radiation models seems required to explain the persistence of CO2 frost in spite of the large apparent energy imbalance during the summer.