Abstract Information on the thermal structure prevailing over the SL9 impact sites hours to months after the collisions is available from measurements of thermal emission at selected wavelengths. Spectral measurements of CH 4 and C 2H 2 emission lines indicate that the stratospheric heating over the impact sites was primarily confined to the pressure levels affected by the fallback of the ejecta plumes ( p<500 μbar for L, p<20 μbar for E). Perturbations amounting to several tens of kelvins around 10 μbar were found over areas 15000–20000 km wide centered on the K and L sites within a day following impact. A large fraction of the kinetic energy of the plumes appeared to have been used to heat the jovian atmosphere rather than being rapidly radiated away. Existing millimeter and thermal infrared measurements give no evidence for a significant heating of the lower stratosphere or upper troposphere, beneath the region of the plume fallbacks. Enhanced brightness observed over many sites in the H 2He continuum very likely results from stratospheric emission of dust particles and not from tropospheric heating. Temperature perturbations from large impactors (G, K, L) were no longer detectable a week after the impacts. Months later, millimeter measurements of shock-produced compounds consistently indicate a temperature of 150–160 K around 100 μbar, likely representative of the nominal state of Jupiter's atmosphere. The E site was still ≈ 40 K warmer than nominal around −3μbar, 2.6 days after impact. Small temperature elevations in this region (≈ 10 K) were still marginally visible over the W and Q 1 sites, 8–10 days after impact. Globally, measurements from various observers consistently indicate that the large sites cooled rapidly with a timescale of 1–2 days, much smaller than the radiative relaxation time in normal quiescent conditions. Enhanced cooling from the gases manufactured and deposited by the impact processes cannot explain this behavior. On the other hand, the mass of silicate particles simultaneously deposited by the plumes is enough to produce the large cooling rate observed. Detailed time-dependent modelling of the radiative balance over the impact sites shows that silicate particles can reproduce the observed evolution of temperatures, provided that their radius is smaller than ≈0.05 μm to delay sedimentation. Mixing with ambient air was probably significant to further reduce the temperature perturbations after several days. A detailed investigation of two particular events, E and H, strongly suggests that the two plumes had vastly different dust-to-gas ratios, a caveat to remember when synthesizing the characteristics of a “generic” event. Further analyses of existing data and further modeling efforts should help refining our view of the perturbations caused by the SL9 collision.