This paper reconsiders the path of the growth of American cities since 1790 (the first census published) in light of new theories of urban growth. Our null hypothesis for long-term growth is random growth, but the alternative is not only mean reversion as is usual. We obtain evidence supporting random growth against the alternative of mean reversion (convergence) in city sizes using panel unit root tests, but we also examine mobility within the size distribution of cities to try to extract growth patterns different from the general unit root trend detected. We find evidence of high mobility when we model growth as a first-order Markov process. Finally, by using a cluster procedure, we find strong evidence in favour of conditional convergence in city growth rates within convergence clubs, which we interpret as “local” mean-reverting behaviours. We interpret the high mobility and the results of the clustering analysis as signs of a sequential city growth pattern toward a random growth steady state.