Dendritic spines represent the major postsynaptic input of excitatory synapses. Loss of spines and changes in their morphology correlate with cognitive impairment in Alzheimer's disease (AD) and are thought to occur early during pathology. Therapeutic intervention at a preclinical stage of AD to modify spine changes might thus be warranted. To follow the development and to potentially interfere with spine changes over time, we established a long term ex vivo model from organotypic cultures of the hippocampus from APP transgenic and control mice. The cultures exhibit spine loss in principal hippocampal neurons, which closely resembles the changes occurring in vivo, and spine morphology progressively changes from mushroom-shaped to stubby. We demonstrate that spine changes are completely reversed within few days after blocking amyloid-β (Aβ) production with the gamma-secretase inhibitor DAPT. We show that the microtubule disrupting drug nocodazole leads to spine loss similar to Aβ expressing cultures and suppresses DAPT-mediated spine recovery in slices from APP transgenic mice. Finally, we report that epothilone D (EpoD) at a subnanomolar concentration, which slightly stabilizes microtubules in model neurons, completely reverses Aβ-induced spine loss and increases thin spine density. Taken together the data indicate that Aβ causes spine changes by microtubule destabilization and that spine recovery requires microtubule polymerization. Moreover, our results suggest that a low, subtoxic concentration of EpoD is sufficient to reduce spine loss during the preclinical stage of AD.