Background A new ARCHITECT^® alpha fetoprotein (AFP) assay was developed to improve the linearity at the upper end of the calibration curve and to enhance other performance characteristics. In addition, this reformulation eliminated the possibility of falsely depressed samples at high AFP concentrations. The purpose of this study was to evaluate its analytical performance at multiple sites. Methods The assay configuration, the diluent formulation, and the manufacturing process were redesigned. Analytical performance was evaluated at Abbott Laboratories, Sapporo Medical University, VU University Medical Center, and Johns Hopkins University. Results The limit of quantitation of the assay was 1.00–1.30 ng/mL. Total precision (%CV) across the assay range varied between 1.41 and 3.52. The assay was linear from 1.19 to 2535 ng/mL, and the range of the assay was expanded from 200 ng/mL to 2000 ng/mL. Comparison of this assay with the on-market ARCHITECT, AxSYM, ADVIA Centaur, DxI, AIA-1800, and E 170 systems yielded regression slopes of 0.91–1.08 and correlation coefficients of ≥0.99 for serum samples. No falsely depressed results were observed in 174 serum samples with AFP concentrations of 2018–1,196,856 ng/mL and in a spiked sample containing up to 10 mg/mL of purified AFP. Conclusions The new AFP assay has improved an issue of the on-market ARCHITECT AFP assay and demonstrated excellent assay performance.