This paper introduces RFCGNN+, a novel learning model that addresses fairness in graph neural networks (GNNs) by incorporating real counterfactual samples and causal analysis. GNNs are powerful tools for processing graph-structured data, but they can inherit and amplify biases present in training data, leading to unfair outcomes for marginalized groups. Existing approaches often focus on sensitive attributes like race or gender as the sole source of bias, but they neglect labeling bias, which arises when societal biases influence data collection. Labeling bias can distort the data, leading to systemic biases in GNNs. To address this, RFCGNN+ uses counterfactual fairness, which aims to eliminate the root causes of inequity by modeling the causal relationships between variables. It identifies authentic counterfactual samples within complex graph structures and incorporates strategies to mitigate labeling bias. Additionally, RFCGNN+ introduces a fairness-aware message-passing framework with multi-frequency aggregation to address topology bias. The model is evaluated on four real-world datasets and a synthetic dataset, demonstrating its effectiveness and practicality. The paper highlights the importance of considering both labeling bias and topology bias in GNNs to achieve fair and equitable outcomes. It also emphasizes the need for further research into the causal mechanisms underlying fairness in graph-structured data. The proposed approach offers a promising direction for developing fair GNNs that can support ethical and equitable decision-making in high-stakes scenarios.This paper introduces RFCGNN+, a novel learning model that addresses fairness in graph neural networks (GNNs) by incorporating real counterfactual samples and causal analysis. GNNs are powerful tools for processing graph-structured data, but they can inherit and amplify biases present in training data, leading to unfair outcomes for marginalized groups. Existing approaches often focus on sensitive attributes like race or gender as the sole source of bias, but they neglect labeling bias, which arises when societal biases influence data collection. Labeling bias can distort the data, leading to systemic biases in GNNs. To address this, RFCGNN+ uses counterfactual fairness, which aims to eliminate the root causes of inequity by modeling the causal relationships between variables. It identifies authentic counterfactual samples within complex graph structures and incorporates strategies to mitigate labeling bias. Additionally, RFCGNN+ introduces a fairness-aware message-passing framework with multi-frequency aggregation to address topology bias. The model is evaluated on four real-world datasets and a synthetic dataset, demonstrating its effectiveness and practicality. The paper highlights the importance of considering both labeling bias and topology bias in GNNs to achieve fair and equitable outcomes. It also emphasizes the need for further research into the causal mechanisms underlying fairness in graph-structured data. The proposed approach offers a promising direction for developing fair GNNs that can support ethical and equitable decision-making in high-stakes scenarios.