Ambient backscatter communication (AmBC) is a newly cutting-edge technology for Internet of Things, which utilizes the ambient radio frequency signal as the carrier to transmit information. Existing works focus on the simple on-off keying modulation which has low channel utilization. However, it is not desirable to develop high-order modulation in the power domain due to the weak strength of the backscatter signal. In this paper, we extend the high-order modulation in the time domain instead, i.e., high-order time shift keying (TSK). We study the performance of the high-order TSK modulation for AmBC for both the complex Gaussian (CG) source signals and M-ary phase shifting keying (MPSK) source signals. The maximum likelihood detector and energy detectors are developed. Particularly, when the training symbols are sent, a better bit-error-rate (BER) performance can be achieved in some cases. When the training symbols are not sent, the receiver can also decode backscattered symbols effectively. Simulation results show that large modulation order is suitable for the MPSK source across all bit signal-to-noise ratio (b-SNR) while is only in the low b-SNR region for the CG source.