Picosecond ultrasonics, as a nondestructive and noncontact method, can be employed for nanoscale metallic film thickness measurements. The sensitivity of the system, which determines the measurement precision and practicability of this technique, is often limited by the weak intensity of the ultrasonic signal. To solve this problem, we investigate the distinct mechanisms involved in picosecond ultrasonic thickness measurement for two types of metals, namely tungsten (W) and gold (Au). For thickness measurement in W films, theory and simulation show that optimizing the pump and probe laser wavelengths, which determine the intensity and shape of the ultrasonic signal, is critical to improving measurement sensitivity, while for Au film measurements, where acoustic-induced beam distortion is dominant, the signal intensity can be optimized by selecting an appropriate aperture size and sample position. The above approaches are validated in experiments. A dual-wavelength pump–probe system is constructed based on a passively mode-locked ytterbium-doped fiber laser. The smoothing method and multipeak Gaussian fitting are employed for the extraction of ultrasonic time-of-flight. Subnanometer measurement precision is achieved in a series of W and Au films with thicknesses of 43–750 nm. This work can be applied to various high-precision, noncontact measurements of metal film thickness in the semiconductor industry.