We studied for the first time the objective function of minimizing total early work. In this setting only early completed jobs are penalized, and the cost of a given job is a function of its processing time. We solved the single machine version of the problem, under the natural assumption of non-delay. We proved that the problem is NP-hard, and introduced a pseudo-polynomial dynamic programming algorithm. Our numerical tests indicate that the dynamic program ming is efficient even for relatively large instances (of up to 200 jobs).