This paper demonstrates, by pulse-chase techniques, the binding to rat liver mitochondrial carbamoyl phosphate synthetase of the ATP molecule (ATPB) which transfers its gamma-phosphoryl group to carbamoyl phosphate. This bound APTB can react with NH3, HCO-3 and ATP (see below) to produce carbamoyl phosphate before it exchanges with free ATP. Mg2+ and N-acetylglutamate, but not NH3 or HCO-3, are required for this binding; the amount bound depends on the concentration of ATP (Kapp = 10--30 microns ATP) and the amount of enzyme. At saturation at least one ATPB molecule binds per enzyme dimer. Binding of ATPB follows a slow exponential time course (t1/2 8--16 s, 22 degrees C), independent of ATP concentration and little affected by NH3, NCO-3 or by incubation of the enzyme with unlabelled ATP prior to the pulse of [gamma-32P]ATP. Formation of carbamoyl phosphate from traces of NH3 and HCO-3 when the enzyme is incubated with ATP follows the kinetics expected if it were generated from the bound ATPB, indicating that the latter is a precursor of carbamoyl phosphate ('Cbm-P precursor') in the normal enzyme reaction. This indicates that the site for ATPB is usually inaccessible to ATP in solution but becomes accessible when the enzyme undergoes a periodical conformational change. Bound ATP becomes Cbm-P precursor when the enzyme reverts to the inaccessible conformation. Pulse-chase experiments in the absence of NH3 and HCO-3 (less than 0.2 mM) also demonstrate binding of ATPA (the molecule which yields Pi in the normal enzyme reaction), as shown by a 'burst' in 32Pi production. Therefore, (in accordance with our previous findings) both ATPA and ATPB can bind simultaneously to the enzyme and react with NH3 and HCO-3 in the chase solution before they can exchange with free ATP. However, at low ATP concentration (18 micron) in the pulse incubation, only ATPB binds since ATP is required in the chase (see above). Despite the presence of two ATP binding sites, the bifunctional inhibitor adenosine(5')pentaphospho(5')adenosine(Ap5A) fails to inhibit the enzyme significantly. A more detailed modification of the scheme previously published [Rubio, V. & Grisolia, S. (1977) Biochemistry, 16, 321--329] is proposed; it is suggested that ATPB gains access to the active centre when the products leave the enzyme and the active centre is in an accessible configuration. The transformation from accessible to inaccessible configuration appears to be part of the normal enzyme reaction and may represent to conformational change postulated by others from steady-state kinetics. The properties of the intermediates also indicate that hydrolysis of ATPA must be largely responsible for the HCO-3-dependent ATPase activity of the enzyme. The lack of inhibition of the enzyme by Ap5A indicates substantial differences between the Escherichia coli and the rat liver synthetase.