To solve the job shop scheduling problem, an improved teaching-learning-based optimization algorithm(TLBO) is proposed in this paper. Aiming at the weak local search ability in the existing TLBO, it is proposed that the learner should learn knowledge not only from the better learners, but also from itself, therefore, a differential self-learning operator is designed. The learning ability of learner is evaluated by its optimal, and its learning times are adaptive calculated according to its learning ability. The learners with higher learning ability have more chance to self-learn. Finally, the proposed method is applied to solve the benchmark instances in OR-Library. The experimental results show that the proposed algorithm is effective while solving the job shop scheduling problem, and its accuracy and robustness can be improved further.