Observer disagreement is important insofar as it limits the reliabilities of observational records. This discussion evolves around methods and conditions under which observer agreement can be measured as to minimize such an occurrence. Observers should be trained to nearly perfect agreement with a criterion or expert coder on unambiguous examples of behavioral categories before actual data collection. Disagreement on ambiguities may help reflect a more accurate representation of the real world. In addition to criterion-related agreement, it is suggested that intraobserver agreement be obtained by showing a video tape twice to all observers in which conditions parallel those encountered in the field. While criterion-related and intraobserver agreement measures have been recommended for both before and during a study, they should not be used as evidence of observer agreement in the actual classroom, but rather to assist an investigator in documenting adequacy of observational skills. After a study is finished, reliabilities of observational data and coefficients of stability and observer agreement should be calculated by using intraclass correlation coefficients. (Author/RC)