Secure two-party computation considers the problem of two parties computing a joint function of their private inputs without revealing anything beyond the output of the computation. In this work, we take the first steps towards understanding the setting in which the two parties want to evaluate a joint quantum functionality while using only a classical channel between them. Our first result indicates that it is in general impossible to realize a two-party quantum functionality against malicious adversaries with black-box simulation, relying only on classical channels. The negative result stems from reducing the existence of a black-box simulator to an extractor for classical proof of quantum knowledge, which in turn leads to violation of the quantum no-cloning. Next, we introduce the notion of oblivious quantum function evaluation (OQFE). An OQFE is a two-party quantum cryptographic primitive with one fully classical party (Alice) whose input is (a classical description of a) quantum unitary, $U$, and a quantum party (Bob) whose input is a quantum state, $\psi$. In particular, Alice receives a classical output corresponding to the measurement of $U(\psi)$ while Bob receives no output. In OQFE, Bob remains oblivious to Alice's input, while Alice learns nothing about $\psi$ more than what can be learned from the output. We present two constructions, one secure against semi-honest parties and the other against malicious parties. Due to the no-go result mentioned above, we consider what is arguably the best possible notion obtainable in our model concerning malicious adversaries: one-sided simulation security. Our protocol relies on the assumption of injective homomorphic trapdoor OWFs, which in turn rely on the LWE problem. As a result, we put forward a first, simple and modular, construction of one-sided quantum two-party computation and quantum oblivious transfer over classical networks.