Femtosecond transient absorption spectra on reaction centers from Rhodobacter sphaeroides wild type have been recorded with high time and wavelength resolution and a very high S/N ratio in the 500-940 nm range with a diode array system. The data have been analyzed by global analysis. Five lifetime components of 1.5, 3.1, 10.8, and 148 ps and long-lived (several nanoseconds) were required to fit the entire three-dimensional data surface adequately with a single set of lifetimes and decay-associated difference spectra (DADS). Up to 30 ps, there is little dispersion in the lifetimes, but in the longer time range (50-250 ps), a substantial variation in lifetime was observed, depending on detection wavelength. The data from the global analysis have been subjected to kinetic modeling comparing sequential kinetic schemes either including (reversible model) or excluding (forward model) back-reactions in the early electron transfer process(es). Thus, the molecular rate constants for the model(s) and the difference spectra of the pure intermediates [species-associated difference spectra (SADS)] were obtained. The data unequivocally confirm the necessity of an electron transfer intermediate with spectral characteristics of P+B-H prior to the formation of the P+BH- state (P is special pair, B is accessory chlorophyll, and H is pheophytin), irrespective of the model chosen. Besides being in much better agreement with the observation of long-lived fluorescence kinetics components, the reversible model results in SADS, in particular for the P+BH- state, that are in somewhat better agreement with expectations than for the pure forward model. For these and other reasons, the reversible model is preferred over the pure forward model. The electrochromic shifts of the H bands in the P+B- state and of the B bands in the P+H- state are revealed clearly in the spectra, thus supporting the assignments. Within the reversible model, the rate constant for the forward reaction in the first step P*-->P+B-H is slightly larger [k12 approximately (2.48 ps)-1] than for the second step P+B-H-->P+BH- [k23 approximately equal to (2.53 ps)-1], in contrast to the pure forward model. From the rate constants for the respective back-reactions, the free energy differences delta G relative to P* for the states P+B-H and P+BH- have been determined to be -41 and -91 meV, respectively. Thus, the free energy difference for the P+BH- state at early times after electron transfer is by a factor of 2-3 smaller than assumed so far. This has the important consequence that a quasi-equilibrium exists from about 10 ps until further electron transfer on the 200 ps time scale with a substantial percentage (approximately 16%) of the P+B-H state present. These results present the first direct evidence from transient absorption data, where the nature of the intermediate can be assigned, for the validity of the slow radical pair relaxation concept. The results have various consequences for understanding the mechanism of the overall electron transfer reaction and imply a much more active role of the protein in the early charge separation processes of the reaction center than assumed so far. The data are discussed in terms of current electron transfer theory. It is suggested that the two first-electron steps operate at a rate very close to the maximal possible rate.