The optimization of a twin-column recycling separation process (TCRSP) for maximum resolution or maximum speed-resolution was investigated. The general optimization method was based on the construction of kinetic plots by assuming an ideal TCRSP (no efficiency loss upon recycling). For the optimization, we examined three chromatographic parameters: operation pressure (3000, 6000, 9000, and 12,000psi), column length (10, 15, and 25cm), and column inner diameter (i.d.) (2.1, 3.0, and 4.6mm). Accordingly, the highest TCRSP resolution level is expected for 25cm long columns packed with 2.5, 2.0, 1.7, and 1.6μm particles at pressures of 3000, 6000, 9000, and 12,000psi, respectively. The maximum speed-resolution performance is expected for 10cm columns packed with 3.7, 3.0, 2.6, and 2.4μm particles. 3.0mm i.d. columns are best to minimize the negative impacts of thermal and inter-column dispersion effects on the TCRSP performance. The method was illustrated for the challenging separation (selectivity factor α<1.02) of small molecules in RPLC at a maximum pressure of 6000psi using commercially available columns. Accordingly, 3.0×150mm columns packed with 2.5μm cellulose-1 Trefoil particles (chiral separation, γ-phenylbutyrolactone, α=1.01, efficiency N=4500) and 2.7μm Cortecs-C18 particles (isotope separation, α=1.02, N=14, 500) particles were found to be the most suitable columns to maximize speed-resolution performance. Further optimization of the TCRSP performance was required by reducing the inter-column sample dispersion that could cause undesirable peak tailing. A standard 2.4μL Rheodyne valve and 100μm i.d. tubes were replaced with a home-made 0.5μL low-dispersion prototype valve and 75μm i.d. perfect connection tubes. As a result, the experimental resolution factors were increased by +60% (γ-phenylbutyrolactone, 25 cycles, Rs=0.7→1.1) and +80% (deuterated benzenes, 22 cycles, Rs=1.1→2.0). Direct comparison between the experimental and the predicted TCRSP performance unambiguously demonstrated that the resolution gain was explained by the significant reduction of the peak tailing after a large number of cycles (n>20).