Monoamine oxidase–B (MAO-B) has recently emerged as a potential therapeutic target for Alzheimer’s disease (AD) because of its association with aberrant γ-aminobutyric acid (GABA) production in reactive astrocytes. Although short-term treatment with irreversible MAO-B inhibitors, such as selegiline, improves cognitive deficits in AD patients, long-term treatments have shown disappointing results. We show that prolonged treatment with selegiline fails to reduce aberrant astrocytic GABA levels and rescue memory impairment in APP/PS1 mice, an animal model of AD, because of increased activity in compensatory genes for a GABA-synthesizing enzyme, diamine oxidase (DAO). We have developed a potent, highly selective, and reversible MAO-B inhibitor, KDS2010 (IC₅₀ = 7.6 nM; 12,500-fold selectivity over MAO-A), which overcomes the disadvantages of the irreversible MAO-B inhibitor. Long-term treatment with KDS2010 does not induce compensatory mechanisms, thereby significantly attenuating increased astrocytic GABA levels and astrogliosis, enhancing synaptic transmission, and rescuing learning and memory impairments in APP/PS1 mice.