Large-scaleobject-oriented(OO)softwaresystemshaverecentlybeenfoundtoshareglobalnetworkcharacteristicssuchassmallworldandscalefree,whichgobeyondthescopeoftraditionalsoftwaremeasurementandassessmentmethodologies.Tomeasurethecomplexityatvariouslevelsofgranularity,namelygraph,class(andobject)andsourcecode,weproposeahierarchicalsetofmetricsintermsofcouplingandcohesion-themostimportantcharacteristicsofsoftware,andanalyzeasampleof12open-sourceOOsoftwaresystemstoempiricallyvalidatetheset.Experimentalresultsofthecorrelationsbetweencross-levelmetricsindicatethatthegraphmeasuresofoursetcomplementtraditionalsoftwaremetricswellfromtheviewpointofnetworkthinking,andprovidemoreeffectiveinformationaboutfault-proneclassesinpractice.