The ISC Bulletin is the most widely used data set for traveltime tomography since it comprises the largest collection of arrival times. The large number of stations and events gives the best ray path coverage of the Earth’s mantle currently available, but the data also contain a considerable amount of noise. The effects of noise are supposedly reduced by averaging delay times for similar paths, choosing a suitable parametrization and damping the tomographic inversions. In this article we focus on two types of errors in body wave arrival times and estimate their effects on global tomographic models using synthetic tests. The first type of error stems from the finite reading precision of arrival times and is equivalent to a round-off error. This yields a random contribution to delay times. We show that the reading precision inferred here often does not coincide with that reported in the Bulletin. The influence on tomography of a reading precision of 1 s or better is almost insignificant since (1) its variance is very small compared to the total variance of ISC delay times and (2) less than 5 per cent of that variance maps into the tomographic model. A few stations report some arrival times with an indicated reading precision of 0.1 s that are in reality only picked to the closest 10 s or 1 min. This results in a S/N ratio much lower than 1 and these data should be removed. The second type of error causes systematic variations of delay times as a function of time. A large amount of this error maps into the tomographic model, but, luckily, the size of the error in the input data is one order of magnitude smaller than the standard deviation of the ISC delay times. A test reveals that the rms amplitude due to these systematic errors is between 4.2 per cent (0–35 km depth) and 14.4 per cent (1800–2000 km depth) of the model rms amplitude. This blurs the tomographic model to some degree but does not change the overall amplitude or shape of seismic anomalies.