1. The paper represents an attempt to estimate the error of the standard 10 ml. Haldane gas analysis apparatus under routine conditions of use.
2. The ultimate error of the instrument under optimum conditions has been obtained from duplicate mercury calibrations of the same instrument and corresponds to s.d. = 0·013% gas. A significant day-to-day variation in calibration was found.
3. The accuracy in routine use has been determined by replicate estimations of CO2 and O2 in air and from estimation of CO2 in designed experiments. The data show an error corresponding to s.d. = 0·027 to 0·13%. This is far larger than that described by Haldane.
4. Examination of the distribution of terminal digits in burette readings shows that the smallest scale division is in fact not being divided into ten parts by the worker. Preference for certain digits varies with the individual, with time, and with the accuracy required.
5. There is a marked tendency for a worker to obtain an ‘expected result’, whether this is derived from a previous estimation of the same sample or from knowledge of the expected result; such bias tends to reduce the apparent error of estimation. Bias of replication—‘prejudice error’—may play a part in many forms of measurement.
6. The results and conclusions are discussed in relation to other published data on errors of measurement in general.