This paper presents a so-called multi-trailer drop-and-pull container drayage (MTDPCD) problem. The MTDPCD problem allows a tractor to carry multiple trailers and a tractor can leave the customer while the container is being packed / unpacked. This mode has been applied in real life and shown its great potential to improve transportation efficiency but not been reported in literature. We model the MTDPCD problem based on node decomposition as a nonlinear program and then linearize it. A backtracking adaptive threshold accepting algorithm is proposed to solve the problem. A named 3-vector procedure is designed to calculate the number of trailers and to check feasibility of solutions. A decoding algorithm including three stages is presented to adjust starting times of trucks given sequences of drayage orders. Experiments based on randomly generated instances validate the aforementioned methodology and indicate that pulling 2 trailers can save about 30% of operating costs than pulling 1 trailer.