The invention discloses a dual recursion based block Markov superposition encoding method. The method includes the following steps: taking an error correction code C[n, k] of which the code length isn and the information bit length is k as a basic code, and encoding an information sequence u of which the length is K=kL into a codeword c of which the length is N=n(L+T); the encoding method furtherincludes the following steps: firstly, dividing the information sequence u of which the length is kL into L equal-length packets u, that is expressed as u=(u(0), u(1),..., u(L-1)), wherein the lengthof each packet is k; for the time t=-1, -2, ..., -m1, setting a sequence w1(t) of which the length is n to a full-zero sequence; for the time t=-1, -2, ..., -m2, setting a sequence w2(t) of which thelength is n to the full-zero sequence; and then, at the time t=0, 1, ..., L-1, transmitting the sequence formula 1 (as shown in the original document) of which the length is k to an encoder ENC of the basic code C[n, k] for encoding, further obtaining a sequence formula 2 (as shown in the original document) of which the length is n, and combining a formula 3 and a formula 4 (as shown in the original document) that are fed back to calculate the t-th sub-sequence c(t) of the codeword c. The scheme of the invention has the advantages of simple encoding, low decoding complexity, channel capacityapproximability and the like, and compared with a traditional block Markov superposition encoding method, the scheme of the invention has a lower decoding error floor and lower decoding complexity.