Counterfactuals are the basis of causal inference in observational studies. The fundamental challenge in counterfactual inference lies in addressing the impact of both observed and unobserved confounders. Despite the multitude of proposed methods aimed at addressing observed confounding bias, these approaches are contingent upon the untestable assumption of Unconfoundedness, which posits the absence of unobserved confounders. In this paper, we present a practical framework of Counterfactual Inference Considering Unobserved Confounders (CIUC), which effectively addresses the influence of both observed and unobserved confounders, thereby enabling precise estimation of counterfactual outcomes. Specifically, the CIUC framework begin by employing variational learning to derive the distribution of unobserved confounders disentangled from observed covariates, relaxing the untestable assumption of Unconfoundedness. Furthermore, within the framework, we incorporate a balanced representation model that takes into account both observed and unobserved confounders. This integration guarantees the attainment of unbiased inferences. The CIUC framework is versatile, as it can accommodate both discrete and continuous treatment variables. Additionally, it seamlessly integrates with various existing counterfactual inference models, making it applicable across a wide range of scenarios. In contrast to the majority of existing methods, the CIUC framework goes beyond by offering confidence intervals for the counterfactual outcomes, which proves highly advantageous for risk-sensitive tasks. Extensive experiments conducted on synthetic, semi-synthetic, and real-world datasets provide compelling evidence of the CIUC’s outstanding performance in generating unobserved confounders, learning balanced representations, and accurately estimating treatment effects at both group and individual levels.