Explore our new Cybersecurity Terms Glossary, explained by our experts.
Cybersecurity Terms Glossary
NetFlow anomaly detection finds suspicious network activity in flow metadata — who talked to whom, for how long, and how much — without ever reading the payload.

What Is NetFlow Anomaly Detection?
A flow record is a summary of one conversation between two addresses: source and destination, ports, protocol, start and end time, packet and byte counts, and the TCP flags seen along the way. It contains no payload. NetFlow anomaly detection is the practice of baselining those summaries and flagging the ones that do not fit, which turns a dataset nobody reads into one of the few monitoring sources that still works when everything on the wire is encrypted.
Cisco created NetFlow; the standardised version is IPFIX, defined in RFC 7011, and sFlow takes a sampled approach to the same idea. Cloud providers emit the same shape of data as VPC flow logs and NSG flow logs, so the techniques below transfer directly to a cloud estate.
Four attack behaviours show up clearly in flow metadata even when the content is opaque.
“Anomalous” means nothing without a reference. Baselines need to be narrow enough to be meaningful and broad enough to be stable, which in practice means several of them at once: per host, per subnet or role group, and per time of day. Business traffic has a strong weekday rhythm, and a model that ignores it will alert every Monday at 09:00 and stay silent at 03:00 on Sunday when it matters.
Simple statistics carry most of the load. A moving percentile of outbound bytes per host, a count of distinct external destinations per hour, and a first-seen list of destination domains or ASNs will detect more real intrusions than an unexplained machine learning score. Keep a record of what a host did last week, not only what it is doing now, because the interesting signal is usually the change.
Flow records are cheap to retain, which makes them the natural long-term layer of your security telemetry. A year of flow metadata occupies a fraction of the storage a year of full packet capture would, and it is often the only record that still exists when an intelligence report arrives naming infrastructure that was active eight months ago. That retrospective search is where flow data earns its keep.
It works best alongside sources that explain the gaps: proxy logs for the URLs behind an outbound session, DNS logs for the name that resolved to the address, and endpoint telemetry for the process that opened the socket. Dedicated network detection and response platforms wrap these sources together, but a team with nothing more than flow exports, a sensible baseline, and a query tool can get a long way.
UK operators of essential services will find network monitoring sitting squarely under Objective C of the NCSC’s Cyber Assessment Framework, which expects organisations to collect and analyse network data well enough to detect intrusions. At Privia Security, we assess and extend monitoring coverage of exactly this kind through our SOC maturation services.
NetFlow records metadata about each conversation: addresses, ports, timings, and volumes. Full packet capture stores the traffic itself. Capture answers questions flow data cannot, such as what was in a request, but it costs orders of magnitude more storage and usually survives days rather than months. Most teams keep a year of flow data and a few days of capture at critical boundaries.
Yes, because the giveaway is the pattern rather than the content. Regular intervals, consistent small payload sizes, a long-lived connection to a destination no other host contacts, and a session that continues outside working hours are all visible in metadata. Encryption hides what is being said, not that something is being said on a schedule.
It breaks some of it. Sampled flow data remains accurate for volumetric questions such as top talkers and sudden bandwidth changes, because the sample scales. It is unreliable for anything that depends on seeing every session, including beaconing with long intervals and single-packet scanning. Collect unsampled data at network edges and accept sampling only on high-throughput core links.
Endpoint tools cover managed devices only. Printers, cameras, building management controllers, network appliances, contractor laptops, and most operational technology will never run an agent, and they are exactly the devices attackers pivot through. Flow data sees them all. The two sources answer different questions, and an intrusion is usually reconstructed from both.
You May Be Interested In These
We use a few cookies that are needed to make this website work, and they are always on. We would also like to set analytics cookies, to see which pages are useful; functional cookies, so that our live chat works; and advertising cookies, to measure our campaigns. We will not set any of those until you say yes, and you can change your mind at any time from the Cookie Settings link in the footer.
Choose the cookies you are happy for us to use. Strictly necessary cookies are always on; everything else is off until you turn it on.
These cookies keep the website working: they hold your security checks, your form submissions and the choice you make here. The site cannot run without them, so they are always on.
These let our live chat open and remember your conversation, and let the newsletter sign-up work. Switch them off and the chat window will not appear.
These tell us which pages are read and how people find us, so that we can improve the site. The reports are about groups of visitors, never about you personally.
These let us measure our advertising and show you our adverts on other websites, such as LinkedIn, Google and Meta. Switch them off and you will still see adverts, but they will be less relevant.