Let m ≤ n ≤ k. An m × n × k 0-1 array is a Latin box if it contains exactly mn ones, and has at most one 1 in each line. As a special case, Latin boxes in which m = n = k are equivalent to Latin squares. Let (m, n, k; p) be the distribution on m × n × k 0-1 arrays where each entry is 1 with probability p, independently of the other entries. The threshold question for Latin squares asks when (n, n, n; p) contains a Latin square with high probability. More generally, when does (m, n, k; p) support a Latin box with high probability? Let > 0. We give an asymptotically tight answer to this question in the special cases where n = k and m ≤ (1 − ) n, and where n = m and k ≥ (1 + ) n. In both cases, the threshold probability is Θ (log (n) ∕n). This implies threshold results for Latin rectangles and proper edge-colorings of K n,n .