In this paper, we formally prove that padding the plaintext with a random bit-string provides the semantic security against chosen plaintext attack (IND-CPA) for the McEliece (and its dual, the Niederreiter's) cryptosystems under the standard assumptions. Such padding has recently been used by Suzuki, Imai and Kobara in the context of RFID security. Our proof relies on the technical result by Katz and Shin from Eurocrypt '05 showing "pseudorandomness" implied by learning parity checks with noise (LPN) problem. We do not need the random oracles as opposed to the known conversions, while the recent ones provide stronger protection (as compared to our scheme) ? against adaptive chosen ciphertext attack (IND-CCA2). In order to show that the padded version of the cryptosystem remains practical, we provide the estimates for suitable key size together with corresponding work required for successful attack.