A cold-cathode gauge does not immediately start when power is applied. Starting is generally assumed to be triggered by a cosmic ray or other energetic particle. The starting behaviors of two magnetron, two inverted magnetron, and five double inverted magnetron gauges were studied at pressures ranging from 2x10(-7) down to 10(-10) Torr. The effects of various starting devices (beta, alpha, field-electron, and photon emitters) were also evaluated. The median starting time for a given gauge was found to be approximately inversely proportional to pressure and consisted of two distinct components : a dormant interval and a growth period. The dormant interval is inherently nonreproducible, being governed by the statistical laws of random events. It can only be quantified by taking the median of a large number of measurements. It is normally by far the larger component and can amount to hours or days at 10(-10) Torr. The growth period, on the other hand, is highly reproducible. It typically has a value of about 1 min at 10(-10) Torr. For most of the growth period the discharge current is below the limit of detection, making it necessary to use special experimental techniques to determine the true growth period. A double inverted magnetron with radioactive triggering to reduce the dormant interval to less than a second at 10(-10) Torr is described. Sources of error in starting-time measurements are discussed.