Molecular communication (MC) via diffusion is envisioned to be a new paradigm for information exchange in the future nanonetworks. However, the strong inter-symbol interference (ISI) caused by the diffusion channel significantly deteriorates the performance of MC systems. To this end, we propose a novel modulation technique to reduce the ISI effect, termed as molecular type permutation shift keying (MTPSK), which encodes information on the permutations of multiple types of molecules. We design a Genie-aided maximum-likelihood detector and a conventional maximum-likelihood detector, and analyze their performance in terms of bit error rate (BER). Aiming at lower computational complexity, we further design a low-complexity maximum-likelihood detector using a Viterbi-like algorithm with compromised error performance. BER simulation results corroborate that the proposed MTPSK can outperform the prevailing modulation schemes for MC, including molecular shift keying (MoSK), concentration shift keying, depleted MoSK, and pulse position modulation.