In this paper we discuss the connection between trajectory and density memory. The first form of memory is a property of a stochastic trajectory, whose stationary correlation function shows that the fluctuation at a given time depends on the earlier fluctuations. The density memory is a property of a collection of trajectories, whose density time evolution is described by a time convoluted equation showing that the density time evolution depends on its past history. We show that the trajectory memory does not necessarily yields density memory, and that density memory might be compatible with the existence of abrupt jumps resetting to zero the system's memory. We focus our attention on a time-convoluted diffusion equation, when the memory kernel is an inverse power law with (i) negative and (ii) positive tail. In case (i) there exist both renewal trajectories and trajectories with memory, compatible with this equation. Case (ii), which has eluded so far a convincing interpretation in terms of trajectories, is shown to be compatible only with trajectory memory. (c) 2007 Elsevier Ltd. All rights reserved.