Axon and synapse degeneration are common components of many neurodegenerative diseases, and their rescue is essential for effective neuroprotection. The chimeric Wallerian degeneration slow protein (Wld(S)) protects axons dose dependently, but its mechanism is still elusive. We recently showed that Wld(S) acts at a non-nuclear location and is present in axons. This and other recent reports support a model in which Wld(S) protects by extranuclear redistribution of its nuclear NMNAT1 portion. However, it remains unclear whether cytoplasmic NMNAT1 acts locally in axons and synapses or at a non-nuclear site within cell bodies. The potency of axon protection by non-nuclear NMNAT1 relative to Wld(S) also needs to be established in vivo. Because the N-terminal portion of Wld(S) (N70) localized to axons, we hypothesized that it mediates the trafficking of the NMNAT1 portion. To test this, we substituted N70 with an axonal targeting peptide derived from amyloid precursor protein, and fused this to NMNAT1 with disrupted nuclear targeting. In transgenic mice, this transformed NMNAT1 from a molecule unable to inhibit Wallerian degeneration, even at high expression levels, into a protein more potent than Wld(S), able to preserve injured axons for several weeks at undetectable expression levels. Preventing NMNAT1 axonal delivery abolished its protective effect. Axonally targeted NMNAT1 localized to vesicular structures, colocalizing with extranuclear Wld(S), and was cotransported at least partially with mitochondria. We conclude that axonal targeting of NMNAT activity is both necessary and sufficient to delay Wallerian degeneration, and that promoting axonal and synaptic delivery greatly enhances the effectiveness.