Individual neurons are very reliable devices, and network computations can be extremely precise with spike precision on the order of 20 ms (e.g. phase precession, STDP). Cognitive computations, such as that underlying memory formation and consolidations, may depend on this reliability and precision to perform complex, large scale computations. However, synapses in hippocampus and neocortex fail to transmit 3 times out of 4. How can such noisy communication channels result in such reliable and precise neural computations? Fellous will first present a biophysical modeling study showing the usefulness of unreliable synaptic transmission in the CA3-CA1 pathway of the hippocampus. He will then introduce a new connectionist model of the hippocampus used to understand how context influences memory consolidation and reconsolidation. He will then speculate on the role of dopamine in memory consolidation and Post-Traumatic Stress Disorder.