The household survey counts people, so someone with two jobs is counted once and the self-employed are included. It is the source of every rate in the employment report: unemployment, labour-force-participation and u6-unemployment.
Its small sample makes it volatile, with a confidence interval on monthly employment change of several hundred thousand. It is also only benchmarked to population estimates once a year, which can produce discontinuities in January that are not real changes.
Example: household employment falls 350,000 in a month while the establishment-survey shows payrolls up 175,000. The divergence is usually sampling noise, and three-month averages reconcile far more often than single months do.
Related: establishment-survey, unemployment-rate, labour-force-participation, nfp, u6-unemployment