Transmitted pressure pulses in ultrasonic B-mode imaging systems are commonly characterized by their center frequency and bandwidth. Both parameters are associated with tradeoffs in spatial resolution and signal-to-noise in ultrasonic system design, with no general understanding of where they are optimal when applied to specific clinical exams. We use the ideal observer and simple psychophysical studies with human observers to evaluate the efficiency of information transfer in B-mode imaging as a function of the transmitted pulse center frequency and fractional bandwidth. Our approach uses a statistical model of backscatter relevant to breast imaging, and a 2-D model of pulse propagation based on Rayleigh-Sommerfeld diffraction theory. The statistics of the backscattered signal are combined in an ideal observer calculation that quantifies the task-relevant information contained in the radio-frequency (RF) signal after delay-and-sum beamforming. This is followed by a psychophysical evaluation of observer performance on B-mode envelope-detected images in three simple tasks. This experimental design allows us to track the flow of diagnostic information through RF acquisition and subsequent reading of the envelope image. In a low-contrast detection task and a high-contrast boundary discrimination task, optimal efficiency for human observers is observed at the highest center frequencies tested (15 MHz) and at moderate bandwidth (40%). For detection of scattering material in a high-contrast hypoechoic lesion, optimal efficiency was observed at lower center frequencies (5 MHz) and higher bandwidth (80%). The ideal observer analysis shows that this task dependence does not arise in the acquisition stage, where efficiency is maximized at 15 MHz with bandwidths of 60% or greater, but rather in the subsequent processing and reading of the envelope image. In addition, at higher frequencies more information is lost in the processing and reading than in the acquisition of reflected signals.