We have characterized the sensitivity and kinetics of a multiplex immunoassay system based on detection of chemiluminescence (CL) at arrays of antibodies. This enzyme-linked immunosorbent assay (ELISA) was based on the spotting of different antibodies in a circular pattern at the bottom of a well of a microtiter plate. Sandwich immunocomplexes within each spot were labeled with horse radish peroxidase, and CL was generated locally to each spot in the array from turnover of luminol substrate. CL from the arrays across the plate was collected in single images; long exposure times were used to maximize sensitivity, and short exposure times were used to extend the dynamic range at higher signals. Image analysis was used to determine the intensity of light from each spot in the array, and intensity was converted to concentration of protein via comparison to a calibration curve. To determine the intrinsic sensitivity of the CL ELISA array, streptavidin horseradish peroxidase (SA-HRP) was captured on an array spotted with biotinylated detection antibodies. The limit of detection (LOD) of SA-HRP was 105 aM, or 3200 enzymes per 50 μL. A single-plex assay for prostate specific antigen (PSA) was developed that had an LOD of 79 aM when the microtiter plate was shaken orbitally, comparable to the most sensitive immunoassays reported to date. Normalization of CL signals in the PSA assay to signal per molecule of SA-HRP showed that the efficiency of the shaken assay was ~40%. When the plates were not shaken, the efficiency was ~4.5%, i.e., ~9-fold lower than when shaken. To better understand the theoretical basis of the sensitivity of these assays, we developed COMSOL numerical models of the binding kinetics at the array for plates that were shaken orbitally and those not shaken. Experimental data from the orbitally shaken PSA assay were best modeled by inertial mixing in a three-layer system that included a 8-μm-thick concentration boundary layer. Experimental data from the unshaken PSA assay were well modeled by diffusion-limited kinetics. A single-plex assay for IL-10 was developed with an LOD of 69 aM or 1.5 fg/mL, and used to measure this cytokine in plasma and serum of 10 healthy individuals. A 5-plex assay for IL-5, IL-6, IL-10, IL-22, and TNF-α was developed with LODs of 56 aM, 237 aM, 69 aM, 88 aM, and 373 aM, respectively. The assay was used to measure these 5 cytokines in the plasma and serum of the same individuals. The correlation in concentration of IL-10 measured in single-plex and multiplex assays was good (r2 = 0.89; bias = 14.5%). The factors that result in the high sensitivity of CL ELISA arrays-mostly high signal to noise ratio of extended chemiluminescent imaging-are discussed. This multiplex CL ELISA could be used for sensitive profiling of multiple proteins for in vitro diagnostics and biomarker detection in the development of therapeutics.