'A circuit' has a response, a behaviour.
That response can be completely described in the time domain, or the frequency domain. It doesn't matter how it is described, it's the same response. You can get from one to the other with the Laplace Transform.
Often, we use just part of the response, and throw away data that's not relevant to our immediate need. When we throw away data, we are then unable to transform that incomplete description into the other one.
For instance, we might take a time domain description and just focus on the percent overshoot and time to 50% of the step response. Or we might take a frequency domain description and use only the magnitude, ignoring the phase. When we throw the rest of the data away, we cannot uniquely reconstruct the time from the frequency, or vice versa.
There are of course some clues in an incomplete response. A large overshoot in the time domain suggests a steep edge in the frequency domain. A narrow passband in the frequency domain would suggest a long time to 50% step response. If the circuits are constrained in topology (so we have some knowledge outside the measured response), then there are some techniques that allow us to do rather better than 'suggesting' a form of response from partial data in the other domain. For instance, for the classic lowpass RC filter (a very simple and constrained circuit), you will often see a relationship given between its 3dB frequency response and its 50% step response.
There may be a good reason for taking data in one domain, when we actually want the other domain. For instance, measuring the impedance versus distance of a long transmission line is called Time Domain Reflectometry. This is often used for diagnosing where incorrect impedances are, due to perhaps faulty connectors, crushing or water ingress. The classic way of doing this was to launch a fast pulse into the line, and see the reflections on a 'scope. However, it's hard work to generate a lot of power in a very short pulse. These days, we get the same information by taking the data in the frequency domain with a network analyser, then transforming the full (amplitude and phase) frequency response measurements into the time domain. Same results, but we can get better frequency resolution and SNR than with a pulse.