In an experimental study of reader experience in identifying the variables essential to grading bone neoplasms, reader error is measured against book grade, a human consensus of the presence or absence of key variables. The average accuracy for classifying focal lesions into slow or fast categories is 83.4% for 890 readings as compared with average diagnostic accuracy of 53.7%. Analyses of human error have provided insight into how to improve the grading algorithm without significant loss of its ability to separate lesions into meaningful categories.