Kluyveromyces marxianus is now considered one of the best choices of option for industrial applications of yeast because the strain is able to grow at high temperature, utilizes various carbon sources, and grows fast. However, the use of K. marxianus as a host for industrial applications is still limited. This limitation is largely due to a lack of knowledge on the characteristics of the promoters since the time and amount of protein expression is strongly dependent on the promoter employed. In this study, four well-known constitutive promoters (P(CYC), P(TEF), P(GPD), and P(ADH)) of Saccharomyces cerevisiae were characterized in K. marxianus in terms of protein expression level and their stochastic behavior. After constructing five URA3-auxotrophic K. marxianus strains and a plasmid vector, four cassettes each comprising one of the promoters--the gene for the green fluorescence protein (GFP)--CYC1 terminator (T(CYC)) were inserted into the vector. GFP expression under the control of each one of the promoters was analyzed by reverse transcription PCR, fluorescence microscopy, and flow cytometer. Using these combined methods, the promoter strength was determined to be in the order of P(GPD) > P(ADH) ∼ P(TEF) >> P(CYC). All promoters except for the P(CYC) exhibited three distinctive populations, including non-expressing cells, weakly expressing cells, and strongly expressing cells. The relative ratios between populations were strongly dependent on the promoter and culture time. Forward scattering was independent of GFP fluorescence intensity, indicating that the different fluorescence intensities were not just due to different cell sizes derived from budding. It also excluded the possibility that the non-expressing cells resulted from plasmid loss because plasmid stability was maintained at almost 100 % over the culture time. The same cassettes, cloned into a single copy plasmid pRS416 and transformed into S. cerevisiae, showed only one population. When the cassettes were integrated into the chromosome, the stochastic behavior was markedly reduced. These combined results imply that the gene expression stochasticity should be overcome in order to use this strain for delicate metabolic engineering, which would require the co-expression of several genes.