Apoptotic cell death is an essential physiological process misregulated in many diseases. A central challenge is to understand the quantitative features of apoptotic regulation, such as all-or-none activation of executioner caspases, and cell-to-cell variability in the time of death. Numerous theoretical models have been proposed to explain these behaviors; there have been few direct experimental tests of such hypotheses. We developed a math model of apoptotic regulation in parallel with extensive experimentation at the single cell level. Using live-cell reporters spanning the caspase activation network to monitor multiple signaling events simultaneously in individual cells, we identified the signal transduction steps in which the graded upstream signal induced by the initial death stimulus is converted into an all-or-none downstream response. Using model reduction and dynamical systems theory, control parameters are identified. Model predictions are verified numerically. Math results led to trials of protein overexpression, siRNA-mediated protein depletion, and pharmacological inhibition, and identified conditions under which all-or-none caspase activation fails. When caspase substrates are only partially cleaved, cells fail to choose decisively between death and survival, creating a potentially dangerous situation in which cells survive despite sustaining significant damage. Our computational and experimental studies have combined to generate a comprehensive model of the caspase regulatory network, which accurately predicts both normal and pathological behavior.