A simple design for Bob in two-way quantum key distribution schemes that employs only passive linear optical components to perform the necessary switching operations is presented. It is shown that, for some specific protocols and/or practical qubit implementations, the losses in Bob's station can be made negligible, hence increasing the maximum distance for secret key generation.Introduction: Quantum key distribution (QKD) is a solution to the problem of generating and sharing a secret key between two distant parties, Alice and Bob, which, under idealised conditions, offers high security by utilising the laws of quantum physics [1]. Recently, novel protocols that employ two-way quantum communication have been proposed: the qubits must make a full round trip between Alice and Bob, who act both as transmitters and as receivers. One class of such protocols, which is the main focus of this Letter, is called semi-quantum key distribution (SQKD), in which one of the parties is classical, whereas the other is quantum [2,3]. In this case, Alice prepares a qubit and then sends it to Bob, who can perform either one of the following operations: (i) do nothing, and reflect the qubit back to Alice in the same state (or in a different state obtained by a fixed unitary transformation); or (ii) measure the qubit in a fixed predetermined basis and resend a new one, corresponding to the measurement result, back to Alice. It is still unknown whether such protocols offer real benefits over the standard ones; however, two-way communication has already been shown to allow completely new paradigms in cryptography, such as counterfactual QKD [4]. In this approach, no information on the secret key is actually transmitted via the quantum channel, and Bob obtains his key from the non-detection events.The simplest and most straightforward way to implement Bob's procedure is using an optical switch. Depending on his choice, he either switches an incoming photon from the quantum channel to (i) a Faraday mirror (FM) or (ii) a measurement apparatus, usually consisting of a polarising beamsplitter (PBS) and two detectors. A third position of the switch could connect to a single-photon source to resend the measured quantum states. This approach, however, introduces two important limitations. The first one is the high insertion loss of the switch. As the qubits must make a full round trip from Alice to Bob and then back to Alice, the quantum channel must be travelled twice, which dramatically increases the link attenuation if compared to the usual one-way QKD schemes. Therefore, all losses must be kept at a minimum level. The second limitation is the switch response time. This can impose restrictions on Alice's maximum pulse rate, which has already achieved the mark of a few gigahertz in recent QKD experiments [5].In this Letter, an alternative switching scheme that relies only on passive, low-loss components is presented. It will be considered that the qubits are codified in the polarisation state of a (pseudo-) singlephoton pulse; encoding in ot...