We present a sampling method for use in small to medium-sized streams that is rigorous and repeatable, and contains the necessary flexibility to be speciesor habitat-specific. Our habitat-based method generates large sample sizes for reliable statistical analyses yet remains cost effective. Streams are mapped, typed, and cross-stream sampling units (belts) are placed to systematically sample each stream based on habitat composition. Biotic and abiotic attributes of the habitat associated with each belt and animal capture are measured or estimated. We employed this sampling design to compare amphibian densities within streams in an old-growth coast redwood (Sequoia sempervirens) ecosystem in northwestern California. Our method targets stream-dwelling larval amphibians, but is also effective for adults and neotenes that share lotic habitats with larvae. This method provides baseline data on amphibian presence and abundance, as well as habitat-specific density estimates. It can also yield highly accurate population estimates for streams where sample sizes are sufficiently large. first signs of environmental stress usually occur at the population level, affecting especially sensitive species (Odum 1992, p 542). Amphibians are potentially excellent bioindicators of both qualitative and quantitative changes in environmental conditions because of their high densities, long life, site fidelity, and marked physiological sensitivity (Baringa 1990; Blaustein and Wake 1990; Vitt et al. 1990; Wake and Morowitz 1990; Wyman 1990; Wake 1991; Blaustein 1994) (but see Pechmann and Wilbur 1994). However, rigorous, repeatable sampling methods are needed to permit investigators to test and exploit the intrinsic advan-tages of amphibians as bioindicators. Stream amphibians are challenging to sample because they are unevenly distributed, cryptically colored, and often hidden within the substrate. Corn and Bury (1989) used an area-constrained search technique for sampling stream amphibians that involved intensive searching of all substrates in a single, 10-m long plot located in a typical stretch of stream. They first walked a few hundred meters of the stream and then selected 1 site that appeared representative of the area they had viewed (purposive selection). However, a single purposively selected sample unit per stream does not allow assessment of variability within streams, and does not allow valid statistical analyses. Welsh (1987) modified Corn and Bury's (1989) method to include 3 systematically spaced, 5-m long sample units per stream, which provided a minimum estimate of variation and allowed a greater choice of analytic approaches. Bury and Corn (1991) later concurred in recommending a minimum of 3 samples per stream. None of these approaches, however, adequately addressed the problem of differential use of stream habitats by amphibian species, and the resulting potential for poor sample estimates. The 2-stage designs proposed by Hankin (1984, 1986) for surveys of stream-dwelling fishes supposed that some unbiased method of estimation could be used to estimate numbers of fish present in natural stream habitat units. In his example applications, Hankin assumed