Let \(M_n\) be the maximal displacement of a branching random walk, where the offspring distribution has finite variance and mean 1 and the increments of the random walk have \((4 + \varepsilon)\)-th finite moment and mean zero. Let \(\beta>0\). The main result is that \(n^{-1/2}M_n\) conditioned on nonextinction till time \(n \beta\) of the branching process converges in distribution as \(n\to\infty\). The long and delicate proof is based on the analysis of the random walk associated to the family tree of a branching process introduced by Harris. The result complements constructions of Aldous (the continuum tree) and Le Gall (superprocesses).