Address Clustering

What Is Address Clustering?

Address clustering (also called wallet clustering or entity resolution) is the process of grouping cryptocurrency addresses that are likely controlled by the same entity based on on-chain evidence and behavioral analysis. The goal is to collapse thousands of pseudonymous addresses into a smaller set of logical wallets, each presumably representing a single user, service, or organization.

Bitcoin transactions are pseudonymous, not anonymous. Every transaction is permanently recorded on the public blockchain, and while addresses are not directly tied to real-world identities, patterns in how those addresses are used can reveal which ones belong together. Clustering exploits these patterns to reconstruct a picture of who controls what.

The technique was pioneered by Reid and Harrigan in 2011 and formalized at scale by Meiklejohn et al. in their 2013 paper "A Fistful of Bitcoins," which demonstrated how simple heuristics could re-identify major Bitcoin services. That paper won the ACM Internet Measurement Conference Test-of-Time Award in 2024, underscoring how foundational this work remains.

How It Works

Address clustering relies on a set of heuristics: rules of thumb derived from how Bitcoin transactions are structured. No single heuristic is perfectly reliable, but when combined, they produce clusters that are useful for investigation and compliance.

Common-Input-Ownership Heuristic

The common-input-ownership heuristic is the most powerful clustering technique. Its logic is straightforward: if two or more addresses appear as inputs in the same transaction, they are presumed to belong to the same entity.

This works because of Bitcoin's UTXO model. To construct a transaction that spends multiple UTXOs, the signer must hold the private keys for every input address. If a transaction consumes UTXOs from addresses A, B, and C, a single party almost certainly controls all three.

The implementation typically uses a union-find (disjoint-set) data structure. Each address starts in its own set. For every transaction with multiple inputs, the sets containing those input addresses are merged:

# Pseudocode: union-find clustering
clusters = UnionFind()

for tx in all_transactions:
    input_addresses = [inp.address for inp in tx.inputs]
    if len(input_addresses) > 1:
        root = input_addresses[0]
        for addr in input_addresses[1:]:
            clusters.union(root, addr)

# After processing, connected components form clusters
wallet_A = clusters.find("1A1zP1...")  # Returns cluster ID

A 2025 study evaluating this heuristic found strong pairwise precision (1.00 on labeled addresses) but significantly lower per-wallet precision (0.36), meaning that while same-entity addresses rarely get split apart, clusters often absorb noise from unlabeled addresses. The heuristic's effectiveness varies dramatically by entity type.

Change Address Detection

When a Bitcoin user spends a UTXO, the entire UTXO must be consumed. If the payment amount is less than the UTXO value, the difference is sent back as change to a new address controlled by the sender. Identifying which output is change extends the cluster by linking sender inputs to the change output.

Several sub-heuristics help identify change outputs:

• One-time address: if a transaction has two outputs and one address has never appeared on-chain before, the new address is likely change
• Address type matching: if all inputs use one script type (e.g., P2WPKH), the output matching that type is likely change, since wallets generate change addresses of the same type
• Round number detection: outputs with round values (multiples of powers of 10) are likely deliberate payment amounts, not change
• Locktime fingerprinting: Bitcoin Core sets locktime to the current block height to prevent fee sniping. If an output is later spent in a transaction with the same locktime behavior, it was likely generated by the same wallet
• Peeling chain detection: when a large UTXO is repeatedly split into a small payment and a large remainder, the larger output continuing the chain is likely change

Behavioral Pattern Analysis

Beyond structural heuristics, analysts examine behavioral signals that suggest shared ownership:

• Transaction timing: submission times reveal timezone and operational hours of an entity
• Fee-rate consistency: repeated use of the same fee rate across transactions suggests the same wallet software or configuration
• Wallet fingerprinting: different wallets produce characteristic transaction structures (input ordering, output ordering, script types, RBF signaling)
• Transaction graph topology: patterns like fan-out (one input, many outputs) and fan-in (many inputs, one output) are characteristic of exchanges and mining pools

How Chain Analysis Firms Use Clustering

Companies like Chainalysis, Elliptic, and TRM Labs build on these heuristics to create comprehensive chain analysis platforms. Their workflow combines multiple evidence layers:

1. Deterministic structural clustering: reproducible, auditable on-chain heuristics (co-spend, change detection) form the foundation
2. Machine-learning signals: pattern recognition flags investigative leads but is not treated as forensic proof
3. Intelligence-based attribution: off-chain data (OSINT, exchange KYC records, law enforcement intelligence) labels clusters with real-world identities at documented confidence levels

These firms maintain massive proprietary databases of labeled addresses (exchanges, darknet markets, ransomware groups, sanctioned entities) and run clustering continuously as new blocks arrive, incrementally updating their union-find structures. Exchanges and financial institutions use these tools for KYC/AML compliance, screening incoming deposits against known illicit clusters in real time.

For a deeper look at tracing techniques and their implications, see the research article on Bitcoin transaction graph privacy defenses.

Use Cases

Law Enforcement and Compliance

Address clustering is the primary tool for tracing illicit fund flows on Bitcoin. Investigators follow UTXO chains through clustered wallets to trace ransomware payments, darknet market proceeds, and sanctioned entity activity, sometimes across 50 or more hops in peeling chains.

Exchange Transaction Monitoring

Exchanges use clustering-powered tools to screen deposits and withdrawals. When a user deposits Bitcoin, the exchange checks whether the sending addresses belong to clusters associated with illicit activity. This is a regulatory requirement under travel rule and anti-money-laundering frameworks.

Tax and Audit

Tax authorities and auditors use clustering to verify that taxpayers have disclosed all their Bitcoin holdings. By clustering known addresses, they can identify undisclosed wallets and unreported transactions, as explored in the crypto payment tax reporting research.

Network Research

Academic researchers use clustering to study Bitcoin's economic structure: how concentrated holdings are, how funds flow between entity types, and how the network evolves over time. Harrigan and Fretter (2016) found that just 1,955 super-clusters accounted for 22% of all Bitcoin addresses.

Countermeasures

Several techniques can break or limit the effectiveness of address clustering. Each addresses different heuristics and comes with its own tradeoffs.

CoinJoin

CoinJoin directly breaks the common-input-ownership heuristic. Multiple independent parties combine their inputs and outputs into a single transaction, making it impossible to determine which inputs correspond to which outputs. Implementations like Wasabi Wallet's WabiSabi protocol and JoinMarket automate this process.

The tradeoff: CoinJoin transactions have a distinctive on-chain fingerprint (multiple equal-denomination outputs), making them identifiable even if the internal mapping is opaque. Some exchanges flag or reject CoinJoin-associated funds.

PayJoin

PayJoin (Pay-to-Endpoint) also breaks the common-input-ownership heuristic, but with an important advantage: PayJoin transactions are indistinguishable from normal payments on-chain. Both sender and receiver contribute inputs, so the assumption that all inputs belong to one party is violated without any visible fingerprint. For more details, see the PayJoin privacy research.

Coin Control

Coin control lets users manually select which UTXOs to spend, preventing the wallet from automatically combining UTXOs with different histories in a single transaction. This does not break existing links but prevents creating new ones. It is especially important for users who receive Bitcoin from multiple sources and want to keep those sources unlinkable.

Avoiding Address Reuse

Address reuse makes clustering trivially easy: every transaction involving a reused address is immediately linked. Modern HD wallets (BIP 32/44) generate a fresh address for every receive by default, but users must avoid manually sharing or reusing addresses.

Silent Payments

Silent Payments (BIP 352) let a receiver publish a single static payment code from which senders derive unique one-time addresses using elliptic-curve Diffie-Hellman. Each payment arrives at a unique address with no on-chain link between them. BIP 352 support is available in Bitcoin Core 28.0+, Cake Wallet, and BitBox02 hardware wallets. For more on Bitcoin's evolving privacy toolkit, see the Bitcoin privacy landscape research.

Risks and Limitations

False Positives and Cluster Collapse

A single false positive (incorrectly merging one address from wallet A into wallet B's cluster) propagates through the entire union-find structure. Every future transaction involving either cluster compounds the error. Researchers call this "cluster collapse": overly aggressive change heuristics cause unrelated wallets to merge into massive super-clusters containing millions of addresses.

Sound methodology favors false negatives over false positives: it is better to under-cluster (miss some links) than to over-cluster (wrongly merge unrelated wallets), especially for law enforcement applications where erroneous attribution has serious consequences.

CoinJoin and Collaborative Transactions

Collaborative transactions like CoinJoin and PayJoin intentionally place inputs from multiple independent parties in one transaction, directly violating the common-input-ownership heuristic. Naive clustering that does not identify and exclude these transactions produces false groupings. Detecting and filtering CoinJoin transactions before clustering is now standard practice, but PayJoin transactions remain invisible to such filters.

Privacy Implications

Address clustering enables powerful surveillance of financial activity. While it serves legitimate compliance and law enforcement purposes, it also reduces the privacy of ordinary users. The tension between financial transparency and individual privacy is a fundamental challenge in Bitcoin's design. Layer 2 solutions like Spark move transactions off the base chain, reducing the on-chain footprint available for clustering analysis.

Evolving Techniques

Both clustering and countermeasures continue to evolve. Graph neural networks (GCN, GraphSAGE, GAT) are being trained on Bitcoin transaction graphs for improved change address detection and address classification. Meanwhile, protocol-level privacy improvements like Silent Payments and Taproot adoption change the on-chain signals that clustering depends on. The graph analysis and taint analysis glossary entries cover related techniques.

This glossary entry is for informational purposes only and does not constitute financial or investment advice. Always do your own research before using any protocol or technology.