1. Spatial patterns in channel morphology and substratum composition at small (1–10 metres) and large scales (1–10 kilometres) were analysed to determine the influence of habitat heterogeneity on the distribution and abundance of larval lamprey.

2. We used a nested sampling design and multiple logistic regression to evaluate spatial heterogeneity in the abundance of larval Pacific lamprey, Lampetra tridentata, and habitat in 30 sites (each composed of twelve 1-m² quadrat samples) distributed throughout a 55-km section of the Middle Fork John Day River, OR, U.SA. Statistical models predicting the relative abundance of larvae both among sites (large scale) and among samples (small scale) were ranked using Akaike's Information Criterion (AIC) to identify the ‘best approximating’ models from a set of a priori candidate models determined from the literature on larval lamprey habitat associations.

3. Stream habitat variables predicted patterns in larval abundance but played different roles at different spatial scales. The abundance of larvae at large scales was positively associated with water depth and open riparian canopy, whereas patchiness in larval occurrence at small scales was associated with low water velocity, channel-unit morphology (pool habitats), and the availability of habitat suitable for burrowing.

4. Habitat variables explained variation in larval abundance at large and small scales, but locational factors, such as longitudinal position (river km) and sample location within the channel unit, explained additional variation in the logistic regression model. The results emphasise the need for spatially explicit analysis, both in examining fish habitat relationships and in developing conservation plans for declining fish populations.