Learning resulting in long-term memory correlates with the formation of new dendritic spines. However, are these newly grown spines only a correlate or a necessity for long-term memory storage? The most direct way to answer this fundamental question is to build a tool to selectively eliminate newly formed dendritic spines, and to then apply this tool in vivo to observe whether a recently acquired long-term memory is destabilized.To this end, we have developed a chemical-genetic selective new spine elimination tool (NSET) in organotypic hippocampal slice cultures. This tool consists of two components: 1. Rapid and time-restricted expression of a tagged actin-stabilizing protein following plasticity induction and 2. Subsequent acute depletion of the tagged actin-stabilizing protein. Using NSET, we can eliminate approximately 75% new spines, compared to 30% on control dendritic branches. Pre-existing spine elimination, however, is comparable to controls. Importantly, structurally potentiated pre-existing spines are neither eliminated nor diminished in size following NSET application.After these in vitro proof-of-concept experiments, we applied NSET in vivo in the mouse basolateral amygdala following auditory fear conditioning. These experiments show that, indeed, long-term auditory fear memory trace is selectively destabilized following NSET application, while relearning and the capacity to store a stable new long-term memory is unaffected. Our first results therefore point to the necessity of new spines in long-term memory storage. Furthermore, NSET holds promise as a new tool to advance our understanding of neural circuit plasticity and information storage in the brain.