A semi-blind channel estimation method and apparatus are provided. The semi-blind channel estimation method includes: step S1: obtaining data that includes a first training sequence and that is received by a receive end; step S2: performing minimum mean square error channel estimation based on the data and the prestored first training sequence, to obtain a channel parameter matrix; step S3: detecting the first training sequence by using a least square detection algorithm, to obtain estimated data; and step S4: using the estimated data as a second training sequence, replacing the first training sequence in step S2 with the second training sequence, and cyclically performing step S2 and step S3 on the second training sequence, until a channel parameter matrix obtained last time is the same as a channel parameter matrix obtained this time, and then stopping circulation, to estimate a final channel parameter matrix.