We have developed a system of differential-output monitors that diagnose current and voltage in the vacuum section of a 20-MA 3-MV pulsed-power accelerator. The system includes 62 gauges: 3 current and 6 voltage monitors that are fielded on each of the accelerator's 4 vacuum-insulator stacks, 6 current monitors on each of the accelerator's 4 outer magnetically insulated transmission lines (MITLs), and 2 current monitors on the accelerator's inner MITL. The inner-MITL monitors are located 6 cm from the axis of the load. Each of the stack and outer-MITL current monitors comprises two separate $B$-dot sensors, each of which consists of four 3-mm-diameter wire loops wound in series. The two sensors are separately located within adjacent cavities machined out of a single piece of copper. The high electrical conductivity of copper minimizes penetration of magnetic flux into the cavity walls, which minimizes changes in the sensitivity of the sensors on the 100-ns time scale of the accelerator's power pulse. A model of flux penetration has been developed and is used to correct (to first order) the $B$-dot signals for the penetration that does occur. The two sensors are designed to produce signals with opposite polarities; hence, each current monitor may be regarded as a single detector with differential outputs. Common-mode-noise rejection is achieved by combining these signals in a $50\mathrm{\text{\ensuremath{-}}}\ensuremath{\Omega}$ balun. The signal cables that connect the $B$-dot monitors to the balun are chosen to provide reasonable bandwidth and acceptable levels of Compton drive in the bremsstrahlung field of the accelerator. A single $50\mathrm{\text{\ensuremath{-}}}\ensuremath{\Omega}$ cable transmits the output signal of each balun to a double-wall screen room, where the signals are attenuated, digitized ($0.5\mathrm{\text{\ensuremath{-}}}\mathrm{ns}/\mathrm{sample}$), numerically compensated for cable losses, and numerically integrated. By contrast, each inner-MITL current monitor contains only a single $B$-dot sensor. These monitors are fielded in opposite-polarity pairs. The two signals from a pair are not combined in a balun; they are instead numerically processed for common-mode-noise rejection after digitization. All the current monitors are calibrated on a 76-cm-diameter axisymmetric radial transmission line that is driven by a 10-kA current pulse. The reference current is measured by a current-viewing resistor (CVR). The stack voltage monitors are also differential-output gauges, consisting of one 1.8-cm-diameter $D$-dot sensor and one null sensor. Hence, each voltage monitor is also a differential detector with two output signals, processed as described above. The voltage monitors are calibrated in situ at 1.5 MV on dedicated accelerator shots with a short-circuit load. Faraday's law of induction is used to generate the reference voltage: currents are obtained from calibrated outer-MITL $B$-dot monitors, and inductances from the system geometry. In this way, both current and voltage measurements are traceable to a single CVR. Dependable and consistent measurements are thus obtained with this system of calibrated diagnostics. On accelerator shots that deliver 22 MA to a low-impedance $z$-pinch load, the peak lineal current densities at the stack, outer-MITL, and inner-MITL monitor locations are 0.5, 1, and $58\text{ }\text{ }\mathrm{MA}/\mathrm{m}$, respectively. On such shots the peak currents measured at these three locations agree to within 1%.