Prediction of metal bioaccumulation and toxicity in aquatic organisms has been based on the free ion activity (e.g., FIAM) or more recently on the binding with the biological/toxicological sites of action (e.g., biotic ligand model). However, metals are bound to various intracellular ligands that may control metal toxicity. In this study, we examined the bioaccumulation, subcellular distribution, and toxicity of Cd in a marine diatom, Thalassiosira nordenskioeldii, under different irradiance levels. The diatoms accumulated more cellular Cd under higher irradiance at low [Cd2+] concentrations, but the accumulation slowed down when the [Cd2+] concentration further increased, implying that cellular binding saturation had been reached. Among the five operationally defined subcellular fractions (metal-rich granule, cellular debris, organelles, heat-denatured protein [HDP], and heat-stable protein [HSP]), Cd was most bound to HSP, whereas it was least bound to HDP. Cd was redistributed with increasing [Cd2+] concentration from the biologically detoxified pool to the presumed metal-sensitive fractions (MSF, a combination of organelles and HDP), which led to higher cellular Cd accumulation, toxicity, and sensitivity. Although diatom growth inhibition was significantly related to [Cd2+] concentration and Cd cellular bioaccumulation, the calculated inhibition concentrations based on MSF or organelles exhibited the least difference, strongly suggesting that MSF can provide the better predictor of Cd toxicity under different irradiance levels compared with [Cd2+] concentration or cellular accumulation. Our results demonstrated that models predicting metal toxicity need to address the subcellular fates of metals and how they respond to external and internal conditions.