ABSTRACT Effective population size ( N e ), census size ( N c ) and their ratio ( N e / N c ) are widely used indicators in conservation genetics, yet long‐term empirical data tracking of all three metrics over long time scales in wild populations remains rare. Using monitoring data from 1995 to 2020, we examined temporal dynamics of N e , N c and their relationship during the demographic collapse of Delta Smelt ( Hypomesus transpacificus ), an endangered fish endemic to the San Francisco Estuary, California, USA. We estimated N e for 21 annual cohorts using restriction site‐associated DNA sequencing data, applying the linkage disequilibrium method, temporal method and a recently developed combined method. Abundance was represented by model‐based survey indices ( N c *). N e varied substantially through time (145–1928), declining in recent years but remaining above commonly cited conservation thresholds for long‐term persistence (e.g., N e > 500) until the final years of collapse. N e was positively correlated with N c * ( r = 0.58), but the shallow slope (β = 0.25) indicates that demographic decline outpaced genetic decline. N e / N c * was consistently low (0.0001–0.056), but it increased significantly through time. Our results show that genetic erosion may lag behind demographic collapse and that N e / N c * is not constant and therefore cannot be easily inferred from either variable alone. This study demonstrates the need for additional empirical data sets to facilitate interpretation of N e , N c * and N e / N c * for the conservation of global biodiversity.