NetFlow is a popular technology in network management nowadays. It supports administrators to get various performance metrics. But metric values estimated from NetFlow records may inaccurate in some circumstance. This paper aims at studying the errors of utilized bandwidth measured by NetFlow. At first, packet streams with a specific format were collected from the backbone by passive measurement and these packet streams are called IP Traces that are used as our experimental data. Then, by using IP Traces, NetFlow records were produced according to NetFlow principles using different sampling ratios and the utilized bandwidths are calculated by NetFlow records at different time granularities. These utilized bandwidths called measured values will be compared to the true values that are directly calculated using IP Traces. At last, the relative errors of utilized bandwidth are presented based on measured values and true values. The experimental results show that bigger time granularity results in smaller error of utilized bandwidth measured by NetFlow, and smaller sampling ratio produces smaller errors. However, if the two factors are considered separately, the error is still large. Only when time granularity is big enough and the sampling ratio is small enough the error could be acceptable. If we want to acquire the satisfying precision, time granularity and sampling ratio need to be set properly. The study results of this paper can provide some guidance and reference for estimating utilized bandwidth with NetFlow.