Various forms of synaptic plasticity, including spike timing-dependent plasticity, can be accounted for by calcium-dependent models of synaptic plasticity. However, recent results in which synaptic plasticity is induced by multi-spike protocols cannot simply be accounted for by linear superposition of plasticity due to spike pairs or by existing calcium-dependent models. In this paper, we show that multi-spike protocols can be accounted for if, in addition to the dynamics of back-propagating action potentials, stochastic synaptic dynamics are taken into account. We show that a stochastic implementation can account for the data better than a deterministic implementation and is also more robust. Our results demonstrate that differences between experimental results obtained in hippocampus and visual cortex can be accounted for by the different synaptic and dendritic dynamics in these two systems.