A small case study of published data on response of dryland wheat to nitrogen fertilizer reveals that replication over space apparently substitutes rather poorly for replication over time, in estimating variance of yield conditional on level of fertilizer.