A new class of low-sensitivity decimation filters based on the polyphase decomposition of infinite impulse response (IIR) transfer functions is proposed. The design approach makes use of optimized positioning of poles and zeros so that transfer functions having unequal numerator and denominator orders are obtained. Comparisons with conventional elliptic filters are shown. An illustrative example of application in passive sonar systems using real noisy data is provided. A weighted stopband filter design is presented and its performance in improving lofargram analysis in passive sonar signal processing is verified.