Measurement bias is a central concept of critical dimension (CD) metrology. Bias is a complex function of sample, tool and time. Bias variation defines the total measurement uncertainty (TMU). TMU is a measure of metrology quality. Precision (or bias variation with time) is only a part of TMU. Often tool-to-tool and sample-to-sample components of bias variation exceed precision. To measure sample-to-sample bias variation, knowledge of the reference CD value for the samples is required. Since bias is sample specific, the technology representative set of calibrated samples has to be created. The described approach has been implemented for a comprehensive evaluation of optical scatterometry (OS) to determine readiness of OS for the 65 nm technology production. The tests covered nine OS applications representative of the technology. The testing revealed that OS metrology is mostly ready to support 65 nm technology production. Both spectral ellipsometry and angular reflectometry OS compete closely on all applications. OS demonstrates acceptable averaged bias for CD and sidewall angle for most applications. Correlation of OS to other metrologies is usually satisfactory. At the same time some problems have been observed. The majority of the tested applications show poor linearity for some measured parameters. Cross-correlation between parameters is usually the cause. OS has trouble to meet the semiconductor industry's tight fleet precision requirements. For all applications, OS tool matching is a major component of fleet precision. The evaluation also exposed some general CD metrology challenges. With accuracy allowance in a sub-nanometer range, it is difficult to find an adequate reference technique to verify and calibrate OS models. Atomic force microscopy (AFM) has been chosen as a reference technique during this evaluation, but it has limitations. Precision, sidewall profile resolution and tip finite dimensions are some of the AFM limitations. OS fleet TMU for many applications is unacceptably high. Further work is needed to better understand the impact of reference data uncertainty on these alarming results. It is clear that to achieve a desired sub-nanometer agreement between reference and OS data, one must pay scrupulous attention to every detail of the experiment.