This paper evaluates the security of the generalized chaotic convolutional coder, which is a recently proposed joint error-correction and encryption scheme integrating the chaotic encryption into the convolutional coding. Our results show that the probability of fully recovering the pseudorandom sequence (PRS) controlling the chaotic switches is at least 0.289 under known-plaintext attack, if the number of available plaintext/ciphertext pairs p is equal to the constraint length k of the chaotic convolutional coder. In the case that p=k+e , where e ∈ Z ⁺ , we prove that the probability to fully deduce the PRS is lower bounded by 1-2 ^-ε . Furthermore, we propose four types of chosen-plaintext attack with different decoding complexities and efficiencies to fully derive the PRS.