Mosquito-borne diseases such as malaria and dengue fever pose a major health problem through much of the world. One approach to disease prevention involves the use of selfish genetic elements to drive disease-refractory genes into wild mosquito populations. Recently engineered synthetic drive systems have provided encouragement for this strategy; but at the same time have been greeted with caution over the concern that transgenes may spread into countries and communities without their consent. Consequently, there is also interest in gene drive systems that, while strong enough to bring about local population replacement, are unable to establish themselves beyond a partially-isolated release site, at least during the testing phase. Here, we develop simple deterministic and stochastic models to compare the confinement properties of a variety of gene drive systems. Our results highlight several systems with desirable features for confinement – a high migration rate required to become established in neighboring populations, and low-frequency persistence in neighboring populations for moderate migration rates. Single-allele underdominance and single-locus engineered underdominance have the strongest confinement properties, but are difficult to engineer and require a high introduction frequency, respectively. Toxin-antidote systems such as Semele, Merea and two-locus engineered underdominance show promising confinement properties and require lower introduction frequencies. Killer-rescue is self-limiting in time, but is able to disperse to significant levels in neighboring populations. We discuss the significance of these results in the context of a phased release of transgenic mosquitoes, and the need for characterization of local ecology prior to a release.