A multi-stage hybrid flow shop problem(MHFSP) is studied with unrelated parallel machines at each stage, where job processing time depends on the assigned machine and the buffer capacity between adjacent production stages is limited. Because it is very difficult to solve this NP-hard problem directly, the original problem is transformed to the MHFSP with blocking and unrelated parallel machines(BMHFSP-UPM). An integer programming model is formulated, and a hybrid heuristic algorithm(HH-GA&TS) is then proposed to handle it based on the genetic algorithm(GA) and tabu search(TS). In this algorithm, a two-dimensional matrix coding scheme based on multi-stage parallel production is designed. A self-adaptive strategy of parameters is then developed according to the formulation of the initial solutions using a two-dimensional matrix cell group. Two single-point inverted crossover procedures using job location and gene location are designed, and a single-point mutation procedure using machine number is introduced, so that the GA is applied to update solutions. Finally, the three moving rules of neighborhood solutions are designed: Sequence exchange of machine number(MNE), exchange of job location(JNE), operation mutation of jobs(JNM). Thus the re-optimization process of TS using MNE-JNE-JNM is completed. Simulation experiments are performed on 720 different sized instances with up to 120 jobs. The results show that the hybrid algorithm can find a better quality schedule compared with the GA, the TS and the NEH-IGA.