The brain can be seen as a self-organized dynamical system that optimizes its information processing and storage capabilities. This is supported by studies across scales, where neuronal activity can show critical dynamics, characterized by the emergence of scale-free statistics as captured, for example, by the sizes and durations of neuronal activity avalanches. Another phenomenon observed during sleep, under anesthesia and in in vitro, is that cortical and hippocampal neuronal networks alternate between “up” and “down” states characterized by very distinct firing rates. Previous theoretical work has been able to relate these two concepts and proposed that up states are critical and down states are subcritical, also indicating that the brain spontaneously transitions between the two. Using high-speed high-resolution calcium imaging recordings of neuronal cultures we were able to analyze the neuronal avalanche statistics in populations of thousands of neurons during “up” and “down” states separately for the first time. We found that both “up” and “down” states can exhibit scale-free behavior when taking into account their intrinsic time scales. In particular, the statistical signature of “down” states is indistinguishable from those observed previously in cultures without “up” states. We show that such behavior can not be explained by network models of non-conservative leaky integrate-and-fire neurons with short-term synaptic depression, even when realistic noise levels, spatial network embeddings and heterogeneous populations are taken into account. Whereas the “up” states can be accurately reproduced by these models.