B-type receptors for the neurotransmitter GABA (gamma-aminobutyric acid) inhibit neuronal activity through G-protein-coupled second-messenger systems, which regulate the release of neurotransmitters and the activity of ion channels and adenylyl cyclase. Physiological and biochemical studies show that there are differences in drug efficiencies at different GABA(B) receptors, so it is expected that GABA(B)-receptor (GABA(B)R) subtypes exist. Two GABA(B)-receptor splice variants have been cloned (GABA(B)R1a and GABA(B)R1b), but native GABA(B) receptors and recombinant receptors showed unexplained differences in agonist-binding potencies. Moreover, the activation of presumed effector ion channels in heterologous cells expressing the recombinant receptors proved difficult. Here we describe a new GABA(B) receptor subtype, GABA(B)R2, which does not bind available GABA(B) antagonists with measurable potency. GABA(B)R1a, GABA(B)R1b and GABA(B)R2 alone do not activate Kir3-type potassium channels efficiently, but co-expression of these receptors yields a robust coupling to activation of Kir3 channels. We provide evidence for the assembly of heteromeric GABA(B) receptors in vivo and show that GABA(B)R2 and GABA(B)R1a/b proteins immunoprecipitate and localize together at dendritic spines. The heteromeric receptor complexes exhibit a significant increase in agonist- and partial-agonist-binding potencies as compared with individual receptors and probably represent the predominant native GABA(B) receptor. Heteromeric assembly among G-protein-coupled receptors has not, to our knowledge, been described before.