An overflow re-separation hydrocyclone was experimentally investigated to reduce coarse-particle contamination in the overflow and improve size classification. Four variables were considered: feed pressure, vortex finder insertion depth, underflow outlet diameter, and connecting tube diameter. The connecting tube diameter mainly controlled the amount of solids entering the overflow separation chamber; increasing this diameter increased the overall classification efficiency. A greater vortex finder insertion depth mainly increased overflow fineness. Enlarging the underflow outlet caused more fine particles to report to the underflow and decreased the overall classification efficiency. Raising the feed pressure produced a finer overflow, but its contribution to the overall efficiency was limited. Based on the single-factor tests, Box–Behnken response-surface models were developed for overflow solids yield and overall classification efficiency, and the two responses were optimized simultaneously. Experimental validation under the selected conditions yielded an overflow solids yield of 24.76% and an overall classification efficiency of 54.21%.