Brain networks possess many non-random features that might present a key to the computational effectiveness and robustness of the nervous system. Information-transmission pathways in neuronal structure, especially in complex connectomes, seem to favor specific recurrent motifs. For intrinsically directional and weighted networks such as those in the brain, it is particularly important to use methods that are sensitive to the weight-encoded information. We use new weighted methods to assess statistics of motives in the connectomes of various animals: \textit{C. elegans}, tadpole, drosophila, mouse. Except for drosophila, all connectomes showed an apparent overabundance of redundancy-enhancing clustering motifs. At the same time, a cyclic motive that can be considered the simplest structure for memory preservation was not more numerous than in the randomized network. We discuss potential relations between these structural patterns and the function of these neuronal circuits. Building on recent weighted and directed clustering methods, we properly define a measure of small-worldness in neuronal networks. In contrast with the consistent over-expression of clustering patterns associated with redundant information transfer, our analysis reveals that small-worldness is not a universal feature of connectomes. It can be related to large distances between some nodes, a probable consequence of modular structures, or low clustering values for a significant fraction of the nodes. On the level of individual neurons, we show how specific motifs single out neurons with a particular function in a learning center of drosophila. We demonstrate that modulatory neurons sending converging information into two compartments within the mushroom body (MB) are at the center of a highly clustered group of neurons. This network feature might be essential for learning sparse encoding of conditioned stimuli. Altogether, our results highlight how the fully-weighted and directional methods can glean information about neuronal circuits.