Motivated by the cell-assemblies theory of the brain, we propose a new formal model of threshold nets (TN). TN are patterned after Petri nets, with a very different firing rule, which removes all tokens upon firing of a transition. The generative power of threshold nets, with and without inhibition, is compared with traditional families of languages. Excitatory TN languages are included by the noncounting regular languages and form an infinite hierarchy for increasing values of threshold. Inhibitory nets are included by the context-sensitive languages. Two new net operators, motivated by the phenomena of growth, learning and brain damage are introduced and compared with Boolean operators.