Background: Measuring CD4 levels is the key laboratory investigation for decision making when initiating HAART, a tertiary prevention measure to reduce HIV/AIDS mortality and morbidity. Inherent biological and analytical variability is common during CD4 enumeration. We cannot control biological variation but how significant is analytical variation to clinical decision making. Objectives: To quantify inter and intra laboratories analytical variation of CD4 counts and percentages and to determine the degree to which time lapse after sample collection contributes to the analytical coefficient of variation (CV%).To estimate the extent of disease misclassification due to CD4 variability if CD4 < 350cells/mm3. Setting: This study was conducted at the HIV clinics of RSSC Hospital, a sugar-cane estate health institution located on the north-east of Swaziland, in Lubombo district, the worst affected by HIV/AIDS in Swaziland. The laboratories involved were Lancet, Good Shepherd (GSH) and National Reference (NRL) laboratories. Study design and method: An analytical diagnostic, cross-sectional (observational) study was used in this study. Using a convenience sampling technique and after obtaining consent from participants, blood was collected in EDTA tubes and sub-divided into three samples, each for Lancet, GSH and NRL. The samples were further split into two at each respective laboratory, one of which was run at 12hours and the other at 24hours from the time of sample collection. Main outcome measures: Student t-test; analytical coefficient of variation (CV%); Bland and Altman (BA) method bias and limits of confidence; BA plots and percentage difference plots; concordance correlation, Pearson and Kappa coefficients; McNemar test for comparison of paired proportion. Results: Fifty three participants consented for participation and of these twenty eight participants were male. The mean CD4 was 373.4 cells/mm3 for Lancet, 395.9 cells/mm3 for NRL and 439.2 cells/mm3 for GSH and subsequent paired t-test revealed some inherent variability. The CV% for CD4 count was 3.5%, 8.4% and 20.1 whilst bias was 7.0, 13.5 and 8.2 for NRL, Lancet and GSH respectively. CD4% had even stronger CV% for all three laboratories. Inter-laboratory bias for Lancet/NRL was -31.5; -64.3 for Lancet/GSH and -38.2 for NRL/GSH at 12hours for CD4 count with only Lancet/GSH having a clinically interchangeable limit of agreement. At 24hours, the trends were similar, possibly confirming stability of CD4 between 12 and 24hours. An assessment of disease misclassification at HAART initiation threshold was performed. The agreement was 81.1% for Lancet/NRL, 88.7% for Lancet/GSH and 77.4% for NRL/GSH corresponding to Kappa values of 0.64, 0.77 and 0.55 respectively. McNemar test for paired proportions revealed that there were no differences between the laboratories when it came to initiating HAART. Conclusions: whilst intra-laboratory variability is minimal, there is some significant inter-laboratory variation of CD4 count and CD4% at the laboratories used in Swaziland. Swaziland should ensure standard SOPs, on -going training and continuous quality improvements for all national laboratories and ensure standards are on par with international recommendations. The national HIV guidelines should possibly enforce two different CD4 counts in decision making to reduce systematic errors. Meanwhile, clinicians should continue to use their clinical judgment in cases of suspicious CD4 count results.