We and our collaborators seek to understand molecular mechanisms of long-term memory in identified elements of a memory-encoding circuit in vivo. Our work on the Drosophila olfactory system has a) outlined a simple neural circuit that encodes habituation memory; b) identified likely components and assembly mechanisms for neuronal ribonucleoprotein (RNP) granules; and c) shown how translational control mechanisms and RNP granules participate in mnemonic processes. Our studies indicate that olfactory habituation arises from the potentiation of inhibitory synapses from a sparse group of local interneurons onto excitatory output neurons in the antennal lobe. The underlying synaptic plasticity mechanism, scaled up from small to large circuits, can create negative images (or inhibitory engrams) of object-encoding cell assemblies and so potentially account for habituation across systems and species. This “negative-image model,” recently supported by observations in the mammalian auditory cortex, explains the key behavioral features of habituation (“gating” and “override”) better than any other current model. I will end by discussing arguments developed in collaboration with colleagues in Oxford, which suggest that inhibitory memory engrams, similar to those involved in habituation, can convert recently encoded memories into latent remote memories that remain accessible to recall, and speculate on possible implications for the function and physiology of sleep, atypical psychiatric states, and dreaming.